Tea quality data safety management monitoring method and system

Abstract

The invention discloses a tea quality data safety management monitoring method and system, and particularly relates to the field of intelligent agriculture, and the method comprises the following steps: I, collecting various tea quality data of tea planting, processing, warehousing and logistics processes in real time, and integrating historical quality evaluation data as a comparison reference source; iI, calculating the reliability weight of each piece of tea quality data in the current environment in real time, performing weighted fusion processing on each piece of data, and performing homomorphic encryption on the fused data; according to the method, privacy leakage is effectively prevented, the adaptability of a detection model is enhanced, the real-time response speed in the detection process is increased, the detection accuracy and the automation level are improved, the reliability and stability of multi-source heterogeneous data fusion are improved, the robustness and the intelligent error correction capability of the whole monitoring system are enhanced, and the monitoring efficiency is improved. And multi-chain linkage, data consistency verification and non-tampering evidence storage are realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent agriculture, and particularly relates to a method and system for secure management and monitoring of tea quality data. Background Art

[0002] Tea production involves multiple links such as planting, picking, processing, packaging, warehousing and logistics. A large amount of structured and unstructured data will be generated in each link, such as environmental sensor data, equipment operation status, tea image features, manual inspection results and historical traceability records, etc. These data have a wide range of sources, different formats, and typical characteristics such as fast dynamic changes, wide spatial distribution, and scattered computing resources. However, in practical applications, the management of tea quality data faces challenges such as difficulties in fusing multi-source heterogeneous data, high risks of data privacy and security, difficulties in data consistency and error correction in distributed scenarios, lack of a credible full-process traceability mechanism, and limited edge computing resources and difficulties in model deployment.

[0003] Existing methods and systems for secure management and monitoring of tea quality data are prone to privacy leakage, and have a slow real-time response speed during the detection process, low detection accuracy and automation level, reducing the reliability and stability of multi-source heterogeneous data fusion. Therefore, we propose a method and system for secure management and monitoring of tea quality data. Summary of the Invention

[0004] The object of the present invention is to solve the problems, and propose a method and system for secure management and monitoring of tea quality data.

[0005] In the first aspect of the implementation of the present invention, a method and system for secure management and monitoring of tea quality data are first proposed. The method includes:

[0006] Ⅰ. Real-time collect various tea quality data in the processes of tea planting, processing, warehousing and logistics, and integrate historical quality assessment data as a comparison reference source;

[0007] Ⅱ. Real-time calculate the reliability weights of various tea quality data in the current environment, perform weighted fusion processing on various data, and perform homomorphic encryption on the fused data;

[0008] Ⅲ. Deploy edge computing devices in tea gardens and primary processing sites, and perform identification and analysis on the appearance, color and shape of tea based on the fused and encrypted tea quality data;

[0009] Ⅳ. Detect abnormal conditions in the process of collecting and processing various tea quality data. If there are abnormalities, trigger multi-level early warnings and perform collaborative optimization;

[0010] V. Upload the tea quality data and the fusion data to the corresponding blockchain for deposit, and construct the tea quality traceability evidence chain through cross-chain data verification.

[0011] As a further solution of the present invention, the tea quality data described in step I specifically includes various data such as temperature and humidity, light, soil composition, pesticide residues, processing temperature control parameters, and logistics status; and each item of tea quality data is collected by each group of devices such as environmental sensors, video image acquisition devices, and production process recording systems.

[0012] As a further solution of the present invention, the specific steps for performing weighted fusion processing on each item of data are as follows:

[0013] S1.1: Collect in real time the tea quality data in different tea processing processes participating in data fusion, and extract the current environmental impact factors. Based on the stability, abnormality rate, update time of the tea quality data, and the consistency with other data sources, construct a confidence score function of the tea quality data under the current environmental conditions.

[0014] S1.2: Through normalization processing, process the confidence scores of each item of tea quality data into weighted coefficients actually participating in the fusion. Then, according to the weighted coefficients of each item of tea quality data currently, dynamically adjust the fusion coefficient. Based on the adjusted fusion coefficient, adjust the weighted coefficients of each item of tea quality data.

[0015] S1.3: According to the weighted coefficients of each item of tea quality data after adjustment, perform weighted fusion on each item of tea quality data to generate fusion data. After the fusion is completed, calculate the estimated standard deviation of the fusion data. If the estimated standard deviation of the fusion data is higher than the preset threshold, trigger an alarm and start manual review.

[0016] As a further solution of the present invention, the specific calculation formula of the confidence score function described in S1.1 is as follows:

[0017] ,

[0018] In the formula, represents the current confidence score of the th item of tea quality data; represents the abnormality rate detected most recently for the th item of tea quality data; represents the freshness of the update time of the th item of tea quality data; represents the consistency score of the th item of tea quality data with the data of other data sources; represents the historical stability score of the th item of tea quality data. , , and represent weight coefficients respectively, where ;

[0019] The specific calculation formula for adjusting the weighted coefficients of each tea quality data described in S1.2 is as follows:

[0020] ,

[0021] ,

[0022] ,

[0023] In the formula, represents the weighted coefficient of the th item of tea quality data; represents the th item of tea quality data; represents the total number of tea quality data; represents the fusion coefficient corresponding to each weighted coefficient; represents the weighted coefficient of the th item of tea quality data after adjustment; represents the flag function indicating whether the th weight is of high confidence;

[0024] The specific calculation formula for weighted fusion described in S1.3 is as follows:

[0025] ,

[0026] In the formula, represents the fusion data at the th moment; represents the observed value of the th item of tea quality data at the moment ;

[0027] The specific calculation formula for estimating the standard deviation described in S1.3 is as follows:

[0028] ,

[0029] In the formula, represents the estimated standard deviation of the fusion data at the th moment; represents the standard deviation of the th item of historical fusion data.

[0030] As a further solution of the present invention, the specific steps for performing homomorphic encryption on the fusion data described in step II are as follows:

[0031] S2.1: According to the preset sensitive information policy table, automatically identify the sensitive fields in the real-time collected tea quality data, mark their encryption attributes, and generate a dedicated public-private key pair for each tea quality data. At the same time, set the ring parameters and modulus space required for encryption;

[0032] S2.2: Adopt an additive lightweight homomorphic encryption algorithm to encrypt and encode the sensitive fields in the marked tea quality data, generate the corresponding ciphertext form, and transmit the encrypted sensitive fields to the corresponding central server and blockchain platform through an encrypted channel based on the TLS + ciphertext verification mechanism, and store them in the form of invisible plaintext;

[0033] S2.3: When data analysis, manual auditing, or visual output is required, decrypt the corresponding sensitive fields at the specified node according to the permission policy.

[0034] As a further solution of the present invention, the specific calculation formula of the additive lightweight homomorphic encryption algorithm described in S2.2 is as follows:

[0035] ,

[0036] In the formula, represents the th original data value the encrypted ciphertext; represents the th original data value; represents the th random perturbation value; represents the th public key.

[0037] As a further solution of the present invention, the specific steps of deploying edge computing devices in the tea garden and primary processing site described in step III are as follows:

[0038] S3.1: Arrange edge computing devices integrating cameras, temperature and humidity sensors, and acquisition control units in the tea gardens and primary processing sites in different regions. At the same time, the edge computing devices in the same region upload the collected groups of tea quality data to a unified central server or the cloud, and each central server or the cloud uses the tea images and environmental data in all the received tea quality data to train a group of high-precision convolutional neural network models;

[0039] S3.2: According to the computing power constraints and memory capacity of each edge computing device, design a lightweight quality detection model adapted to the corresponding edge computing device, and initialize the weights of each constructed lightweight quality detection model by using He initialization or Xavier initialization. Collect the preprocessed historical tea leaf images and environmental data of the corresponding edge computing device, the true quality labels corresponding to each data, and use the predicted probability distribution of the tea leaf images by the high-precision convolutional neural network model as the soft labels;

[0040] S3.3: Divide the preprocessed historical tea leaf images and environmental data into a training set, a test set, and a validation set. Then divide the training set into multiple batches of mini-batch training subsets, and set the completion of all batches of training subsets as one cycle. After that, sequentially input the training subsets into the lightweight quality detection model;

[0041] S3.4: The input layer of the lightweight quality detection model receives each batch of training subsets and sequentially inputs them into the convolutional layer. The convolutional layer establishes a sliding window of a preset size. Then the sliding window starts from the head of each tea leaf image in the training subset and slides according to the preset step size. After each slide, perform element-wise multiplication and summation with the local region image selected by the sliding window using a convolutional kernel of size k×k, and use the calculation result after each slide as a set of output pixel values to form an output feature map;

[0042] S3.5: Then perform average pooling on the output feature map generated by the convolutional layer through the pooling layer to generate a reduced output feature map. Then, from shallow to deep, alternately perform convolutional and pooling operations repeatedly multiple times until the final output feature map is output by the last pooling layer. The fully connected layer receives the output feature map flattened into a vector form, performs weighted fusion based on the preset weight matrix in the fully connected layer, and normalizes the fusion result into a probability distribution through the Softmax layer;

[0043] S3.6: Based on the true quality labels and the soft labels, calculate the total loss value of the lightweight quality detection model through the joint loss function, and input the total loss value into the output layer of the lightweight quality detection model. Based on the chain rule, calculate the gradient value of the total loss value with respect to each layer of the lightweight quality detection model through the backpropagation algorithm, and use the Adam optimizer to update the parameters of each layer of the lightweight quality detection model layer by layer;

[0044] S3.7: After each round of training, the validation set is input into the trained lightweight quality detection model, and the performance of the lightweight quality detection model is evaluated. If the model performance does not reach the preset threshold, the training set is reused to train and optimize the parameters of the lightweight quality detection model, and the training verification is repeated until the model performance reaches the preset threshold. The lightweight quality detection model after training is then tested using the test set, and the lightweight quality detection model that meets the preset detection requirements is deployed to the corresponding edge computing device.

[0045] As a further solution of the present invention, the specific steps of each central server or cloud in S3.1 using the received large amount of tea quality data to train a set of high-precision convolutional neural network models are as follows:

[0046] P1.1: Collect tea image data from different tea gardens and processing workshops under multiple angles and lighting conditions, and simultaneously collect temperature, humidity, and lighting environment data during image acquisition to form a multimodal training data set. Annotate each tea image data in the multimodal data set manually or semi-automatically to establish a classification label set.

[0047] P1.2: Perform various enhancement operations such as rotation, scaling, color perturbation, and contrast adjustment on each tea image data, and perform pixel normalization on each enhanced tea image data. A high-precision convolutional neural network model including an output layer, multiple convolutional layers, activation functions, pooling layers, fully connected layers, and an output layer is constructed and stored on a central server or cloud.

[0048] P1.3: Combine the feature vectors of each group of tea image data in the classification label set with their corresponding environmental features to form joint input data, then input each joint input data into a high-precision convolutional neural network model, build a multi-class classification head, pass the joint input data layer by layer through the forward propagation algorithm, and output the predicted category distribution of each tea image data;

[0049] P1.4: Use the cross entropy loss function to calculate the loss value between the predicted distribution and the true label, and use the backpropagation algorithm and SDG optimizer to optimize the parameters of each network layer of the high-precision convolutional neural network model. Use unused joint input data to evaluate the accuracy, recall rate and F1 score performance indicators of the trained high-precision convolutional neural network model, and adjust the learning rate, convolution kernel size and network depth hyperparameters based on the evaluation results. Repeat the training and verification until the performance indicators of the high-precision convolutional neural network model meet the preset threshold, stop the training, and save the trained model structure and weights in a standard format.

[0050] As a further solution of the present invention, the lightweight quality detection model described in S3.2 can be designed in architecture by using depthwise separable convolution, controlling the number of convolutional layers, the number of channels and the size of convolutional kernels, and adopting activation functions such as ReLU6 or Hard-Swish; and the lightweight quality detection model includes an input layer, a convolutional layer, an activation layer, a pooling layer, a fully connected layer, a Softmax layer, and an output layer.

[0051] In the second aspect of the implementation of the present invention, a tea quality data security management and monitoring system is proposed, including: a collection and preprocessing module, a credibility evaluation module, a data fusion module, a data encryption module, an edge detection module, a detection and early warning module, a collaborative optimization module, and a data archiving module;

[0052] The collection and preprocessing module is used to collect various quality indicators of environmental data, image data, and processing technology data, and preprocess the collected raw data;

[0053] The credibility evaluation module dynamically calculates the credibility scores of the collected data according to the stability, historical performance, and real-time status of the collected data after preprocessing;

[0054] The data fusion module fuses the collected data according to the credibility scores of the collected data;

[0055] The data encryption module is used to perform homomorphic encryption processing on sensitive information in the collected data;

[0056] The edge detection module is used to detect and identify tea images;

[0057] The detection and early warning module is used to detect situations such as abnormal fluctuations in quality data, sensor failures, and data offsets, and trigger multi-level early warnings according to the detection results;

[0058] The collaborative optimization module is used for joint training between multiple tea plantations and processing units, and collaboratively optimizes the parameters of the detection and early warning module;

[0059] The data archiving module generates a digest of each tea quality data and archives it on the chain to build a traceable path.

[0060] As a further solution of the present invention, the specific steps for the data archiving module to build a traceable path are as follows:

[0061] S4.1: Structurally process the tea quality data from different links, generate a unique identifier and a hash digest for each tea quality data, write the hash digests of different stages into the planting chain, the processing chain, and the detection chain respectively, and control the operation of each chain through different central servers or the cloud;

[0062] S4.2: When performing data comparison or traceability between different chains, a verification request is initiated, and cross-chain communication is carried out using a relay contract or a light client. At the same time, a zero-knowledge circuit is constructed between the chains to prove the consistency of the on-chain data. After the proof passes, only the zero-knowledge proofs of "verification successful" and "hash consistency" are submitted.

[0063] S4.3: Another chain for data comparison or traceability receives the zero-knowledge proof of the cross-chain verification request and verifies it through a zero-knowledge verification engine. When the verification is successful, a verification record is automatically generated on the target chain, and the on-chain evidence and cross-chain verification results at each stage are collected in real time and integrated into a complete verifiable tea quality traceability path.

[0064] Advantages of the present invention:

[0065] The present invention proposes a tea quality data security management and monitoring method and system. Through a sensitive field identification strategy and a lightweight homomorphic encryption algorithm, the privacy information in various tea quality data is encrypted and encoded, and transmitted and stored through an encrypted channel. Subsequently, combined with the image and environmental data collected by edge computing devices, a high-precision convolutional neural network model is trained in the cloud, and the high-precision convolutional neural network model is compressed into a lightweight model and deployed to low-computing-power edge devices. During the training phase, the model performance is optimized through forward propagation, backward propagation, and joint loss. Finally, various data are structurally processed to generate unique identifiers and hash digests, which are respectively uploaded to the planting chain, processing chain, and detection chain; through cross-chain communication and zero-knowledge proof mechanism, data consistency verification between heterogeneous chains is realized, a complete and credible tea quality traceability path is constructed, effectively preventing privacy leakage, enhancing the adaptability of the detection model, improving the real-time response speed during the detection process, enhancing the detection accuracy and automation level, improving the reliability and stability of multi-source heterogeneous data fusion, enhancing the robustness and intelligent error correction ability of the overall monitoring system, and realizing multi-chain linkage, data consistency verification, and non-tamperable evidence storage. Description of the Drawings

[0066] The following further describes the present invention with reference to the accompanying drawings.

[0067] Figure 1 It is a flowchart of a tea quality data security management and monitoring method provided by an embodiment of the present invention;

[0068] Figure 2 It is a framework diagram of a tea quality data security management and monitoring system provided by an embodiment of the present invention. Detailed Embodiments

[0069] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0070] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0071] The embodiments of the present invention provide a method and system for the secure management and monitoring of tea quality data. Refer to Figure 1 , Figure 1 which is a flowchart of a method for the secure management and monitoring of tea quality data provided by the embodiments of the present invention. The method includes the following steps:

[0072] Collect various tea quality data in the processes of tea planting, processing, warehousing, and logistics in real time, and integrate historical quality assessment data as a comparison reference source.

[0073] It should be further noted that the tea quality data specifically includes various data such as temperature and humidity, light, soil composition, pesticide residues, processing temperature control parameters, and logistics status; and each item of tea quality data is collected by groups of devices such as environmental sensors, video image acquisition devices, and production process recording systems.

[0074] Calculate the reliability weights of various tea quality data in the current environment in real time, perform weighted fusion processing on the data, and perform homomorphic encryption on the fused data.

[0075] Specifically, collect various tea quality data in different tea processing processes participating in data fusion in real time, extract the current environmental impact factors, and construct a confidence score function of tea quality data under the current environmental conditions based on the stability, abnormality rate, update time of tea quality data, and consistency with other data sources. Through normalization processing, the confidence scores of various tea quality data are processed into weighted coefficients actually participating in fusion. Then, according to the weighted coefficients of various tea quality data currently, dynamically adjust the fusion coefficient. Based on the adjusted fusion coefficient, adjust the weighted coefficients of various tea quality data. According to the adjusted weighted coefficients of various tea quality data, perform weighted fusion on various tea quality data to generate fused data. After the fusion is completed, calculate the estimated standard deviation of the fused data. If the estimated standard deviation of the fused data is higher than the preset threshold, trigger an alarm and start manual review.

[0076] Specifically, according to the preset sensitive information policy table, sensitive fields in the real-time collected tea quality data are automatically identified and their encryption attributes are marked. At the same time, a dedicated public key-private key pair is generated for each tea quality data. Additionally, the ring parameters and modulus space required for encryption are set, and the additive lightweight homomorphic encryption algorithm is used to encrypt and encode the sensitive fields in the marked tea quality data, generating corresponding ciphertext forms. Through an encrypted channel based on the TLS + ciphertext verification mechanism, the encrypted sensitive fields are transmitted to the corresponding central server and blockchain platform and stored in an invisible plaintext form. When data analysis, manual auditing, or visual output is required, the corresponding sensitive fields are decrypted at the specified node according to the permission policy.

[0077] In addition, in this embodiment, it should be noted that the specific calculation formula of the confidence score function is as follows:

[0078] ,

[0079] In the formula, represents the current confidence score of the th item of tea quality data; represents the abnormal rate detected most recently for the th item of tea quality data; represents the freshness of the update time of the th item of tea quality data; represents the consistency score of the th item of tea quality data with data from other data sources; represents the historical stability score of the th item of tea quality data; , , and respectively represent weight coefficients, where ;

[0080] The specific calculation formula for adjusting the weighted coefficients of each tea quality data is as follows:

[0081] ,

[0082] ,

[0083] ,

[0084] In the formula, represents the weighted coefficient of the th item of tea quality data; represents the th item of tea quality data; represents the total number of tea quality data; represents the fusion coefficient corresponding to each weighting coefficient; represents the weighting coefficient of the th item of the adjusted tea quality data; represents the flag function indicating whether the

[0085] The specific calculation formula for weighted fusion is as follows:

[0086] ,

[0087] In the formula, represents the fusion data at the th moment; represents the th item of the tea quality data at the moment ;

[0088] The specific calculation formula for estimating the standard deviation is as follows:

[0089] ,

[0090] In the formula, represents the estimated standard deviation of the fusion data at the th moment; represents the standard deviation of the th item of historical fusion data;

[0091] The specific calculation formula for the additive lightweight homomorphic encryption algorithm is as follows:

[0092] ,

[0093] In the formula, represents the th original data value after encryption; represents the th original data value; represents the th random perturbation value; represents the th public key.

[0094] Deploy edge computing devices in the tea garden and at the primary processing site, and based on the fused and encrypted tea quality data, identify and analyze the appearance, color, and shape of the tea.

[0095] Specifically, edge computing devices integrating cameras, temperature and humidity sensors, and acquisition and control units are arranged in tea gardens and primary processing sites in different regions. At the same time, the edge computing devices in the same region upload the collected groups of tea quality data to a unified central server or the cloud. Each central server or the cloud uses the tea images and environmental data in all the received tea quality data to train a group of high-precision convolutional neural network models. According to the computing power constraints and memory capacity of each edge computing device, a lightweight quality detection model adapted to the corresponding edge computing device is designed. The weights of each constructed lightweight quality detection model are initially assigned in the way of He initialization or Xavier initialization. The historical tea images and environmental data preprocessed by the corresponding edge computing device, and the true quality labels corresponding to each data are collected. The predicted probability distribution of the tea images corresponding to the high-precision convolutional neural network model is used as a soft label. The preprocessed historical tea images and environmental data are divided into a training set, a test set, and a validation set. Then the training set is divided into multiple batches of small-batch training subsets, and it is set that the completion of the use of all batches of training subsets is one cycle. After that, the training subsets are sequentially input into the lightweight quality detection model. The input layer of the lightweight quality detection model receives each batch of training subsets and sequentially inputs them into the convolutional layer. The convolutional layer establishes a sliding window of a preset size. Then the sliding window starts from the head of each tea image in the training subset and slides according to the preset step size. After each slide, the local region image selected by the sliding window is multiplied element by element and summed with a convolutional kernel of size k×k, and the calculation result after each slide is used as a group of output pixel values to form an output feature map. Then, the output feature map generated by the convolutional layer is subjected to average pooling processing through the pooling layer to generate a reduced output feature map. Then, from the shallow layer to the deep layer, the convolutional and pooling operations are repeatedly alternated multiple times until the final output feature map is output by the last pooling layer. The fully connected layer receives the output feature map flattened into a vector form, performs weighted fusion based on the preset weight matrix in the fully connected layer, and normalizes the fusion result into a probability distribution through the Softmax layer. Based on the true quality label and the soft label, the total loss value of the lightweight quality detection model is calculated through the joint loss function, and the total loss value is input from the output layer of the lightweight quality detection model. Based on the chain rule, the gradient value of the total loss value with respect to each layer of the lightweight quality detection model is calculated through the backpropagation algorithm, and the parameters of each layer of the lightweight quality detection model are updated layer by layer using the Adam optimizer. After each round of cycle training, the validation set is input into the trained lightweight quality detection model, and the performance of the lightweight quality detection model is evaluated. If the model performance does not reach the preset threshold, the parameters of the lightweight quality detection model are retrained and optimized using the training set, and the training and validation are repeated until the model performance reaches the preset threshold. After that, the test set is used to test the trained lightweight quality detection model, and the lightweight quality detection model that meets the preset detection requirements is deployed to the corresponding edge computing device.

[0096] It should be further noted that tea leaf image data under multiple angles and lighting conditions are collected from different tea gardens and processing workshops. At the same time, environmental data such as temperature, humidity, and lighting during image collection are synchronously collected to form a multi-modal training data set. The tea leaf image data in the multi-modal data set are labeled manually or semi-automatically to establish a classification label set. Each tea leaf image data is subjected to enhancement operations such as rotation, scaling, color perturbation, and contrast adjustment. After that, pixel normalization processing is performed on the enhanced tea leaf image data. A high-precision convolutional neural network model including an output layer, multiple convolutional layers, activation functions, pooling layers, fully connected layers, and an output layer is constructed and stored in a central server or the cloud. The feature vectors of each group of tea leaf image data in the classification label set are combined with their corresponding environmental features to form joint input data. Then, each joint input data is input into the high-precision convolutional neural network model to construct a multi-class classification head. The joint input data is processed layer by layer through the forward propagation algorithm, and the predicted class distribution of each tea leaf image data is output. The cross-entropy loss function is used to calculate the loss value between the predicted distribution and the true label. The backpropagation algorithm and the SDG optimizer are used to optimize the parameters of each network layer of the high-precision convolutional neural network model. The accuracy, recall rate, and F1 score of the trained high-precision convolutional neural network model are evaluated using the unused joint input data. Based on the evaluation results, hyperparameters such as the learning rate, convolutional kernel size, and network depth are adjusted. Then, training and validation are repeated until the performance indicators of the high-precision convolutional neural network model meet the preset threshold, and then the training is stopped. The trained model structure and weights are saved in a standard format.

[0097] In addition, it should be noted that the lightweight quality detection model can be designed by using depthwise separable convolutions, controlling the number of convolutional layers, the number of channels and the convolutional kernel size, and adopting activation functions such as ReLU6 or Hard-Swish. And the lightweight quality detection model includes an input layer, a convolutional layer, an activation layer, a pooling layer, a fully connected layer, a Softmax layer, and an output layer.

[0098] Detect abnormal conditions during the collection and processing of various tea leaf quality data. If there are abnormalities, trigger multi-level warnings and perform collaborative optimization.

[0099] Upload various tea leaf quality data and fusion data to the corresponding blockchain for deposit, and construct a tea leaf quality traceability evidence chain through cross-chain data verification.

[0100] Based on the same inventive concept, the embodiment of the present invention also provides a tea leaf quality data security management and monitoring system. See Figure 2 , Figure 2Schematic diagram of the structure of a tea quality data security management and monitoring system provided by an embodiment of the present invention, including: a collection and preprocessing module, a credibility evaluation module, a data fusion module, a data encryption module, an edge detection module, a detection and early warning module, a collaborative optimization module, and a data storage and certification module;

[0101] The collection and preprocessing module is used to collect quality indicators of environmental data, image data, and processing technology data, and preprocess the collected raw data; the credibility evaluation module dynamically calculates the credibility scores of the collected data according to the stability, historical performance, and real-time status of the preprocessed collected data.

[0102] The data fusion module fuses the collected data based on the credibility scores of the collected data; the data encryption module is used to perform homomorphic encryption processing on sensitive information in the collected data.

[0103] The edge detection module is used to detect and identify tea images; the detection and early warning module is used to detect situations such as abnormal fluctuations in quality data, sensor failures, and data offsets, and trigger multi-level early warnings based on the detection results; the collaborative optimization module is used to conduct joint training between multiple tea gardens and processing units, and jointly optimize the parameters of the detection and early warning module.

[0104] The data storage and certification module generates a digest of each tea quality data and stores it on the chain for certification, and constructs a traceable path.

[0105] Specifically, structurally process the tea quality data from different links, generate a unique identifier and a hash digest for each tea quality data, write the hash digests of different stages into the planting chain, processing chain, and detection chain respectively, control the operation of each chain through different central servers or the cloud. When performing data comparison or traceability between different chains, a verification request is initiated, and cross-chain communication is carried out using a relay contract or a light client. At the same time, a zero-knowledge circuit is constructed between the chains to prove the consistency of the data on the chain. After the proof passes, only the zero-knowledge proofs of "verification successful" and "hash consistency" are submitted. Another chain that performs data comparison or traceability receives the zero-knowledge proof of the cross-chain verification request and verifies it through a zero-knowledge verification engine. When the verification is successful, a verification record is automatically generated on the target chain, and the on-chain storage and cross-chain verification results of each stage are collected in real time and integrated into a complete verifiable tea quality traceability path.

[0106] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.

Claims

1. A method for monitoring the security management of tea quality data, characterized in that, It includes the following steps: Ⅰ. Collect various tea quality data in the processes of tea planting, processing, warehousing and logistics in real time, and integrate historical quality assessment data as a comparison reference source; Ⅱ. Calculate the reliability weights of various tea quality data in the current environment in real time, perform weighted fusion processing on the data, and perform homomorphic encryption on the fused data; Ⅲ. Deploy edge computing devices in tea gardens and primary processing sites, and based on the fused and encrypted tea quality data, identify and analyze the appearance, color and shape of tea; Ⅳ. Detect abnormal conditions in the process of collecting and processing various tea quality data. If there are abnormalities, trigger multi-level warnings and perform collaborative optimization; Ⅴ. Upload various tea quality data and fused data to the corresponding blockchain for deposit, and build a tea quality traceability evidence chain through cross-chain data verification.

2. The method for monitoring the security management of tea quality data according to claim 1, characterized in that, The specific steps of the weighted fusion processing of the data described in step Ⅱ are as follows: S1.1: Collect various tea quality data in different tea processing processes participating in data fusion in real time, extract the current environmental impact factors, and construct a confidence score function of tea quality data under the current environmental conditions based on the stability, abnormal rate, update time of tea quality data and the consistency with other data sources; S1.2: Through normalization processing, process the confidence scores of various tea quality data into weighted coefficients actually participating in the fusion. Then, according to the weighted coefficients of the current tea quality data, dynamically adjust the fusion coefficient, and based on the adjusted fusion coefficient, adjust the weighted coefficients of various tea quality data; S1.3: According to the adjusted weighted coefficients of various tea quality data, perform weighted fusion on various tea quality data to generate fused data. After the fusion is completed, calculate the estimated standard deviation of the fused data. If the estimated standard deviation of the fused data is higher than the preset threshold, trigger a warning and start manual review.

3. The method for monitoring the security management of tea quality data according to claim 2, characterized in that, The specific steps of the homomorphic encryption of the fused data described in step Ⅱ are as follows: S2.1: According to the preset sensitive information policy table, automatically identify the sensitive fields in the various tea quality data collected in real time, mark their encryption attributes, and generate exclusive public-private key pairs for each tea quality data. At the same time, set the ring parameters and modulus space required for encryption; S2.2: Adopt an additive lightweight homomorphic encryption algorithm to encrypt and encode the sensitive fields in the marked various tea quality data, generate the corresponding ciphertext form, and transmit the encrypted sensitive fields to the corresponding central server and blockchain platform through an encrypted channel based on the TLS + ciphertext verification mechanism, and store them in the form of invisible plaintext; S2.3: When data analysis, manual audit or visual output is required, decrypt the corresponding sensitive fields at the specified node according to the permission policy.

4. The method for monitoring the security management of tea quality data according to claim 1, characterized in that, The specific steps of deploying edge computing devices in tea gardens and primary processing sites described in step Ⅲ are as follows: S3.1: Deploy edge computing devices integrated with cameras, temperature and humidity sensors, and acquisition and control units at tea plantations and primary processing sites in different regions. At the same time, the edge computing devices in the same region upload the collected groups of tea quality data to a unified central server or the cloud. Each central server or the cloud uses the tea images and environmental data in all the received tea quality data to train a group of high-precision convolutional neural network models. S3.2: Design lightweight quality detection models adapted to the corresponding edge computing devices according to the computing power constraints and memory capacities of each edge computing device. Initialize the weights of each constructed lightweight quality detection model using He initialization or Xavier initialization. Collect the preprocessed historical tea images and environmental data corresponding to the edge computing devices, the true quality labels corresponding to the data, and use the predicted probability distribution of the tea images corresponding to the high-precision convolutional neural network model as soft labels. S3.3: Divide the preprocessed historical tea images and environmental data into a training set, a test set, and a validation set. Then divide the training set into multiple batches of small training subsets, and set the completion of all batches of training subsets as one cycle. Then sequentially input the training subsets into the lightweight quality detection model. S3.4: The input layer of the lightweight quality detection model receives each batch of training subsets and sequentially inputs them into the convolutional layer. The convolutional layer creates a sliding window of a preset size. Then the sliding window starts from the head of each tea image in the training subset and slides according to the preset step size. After each slide, perform element-wise multiplication and summation with the local region image selected by the sliding window using a convolutional kernel of size k×k, and use the calculation result after each slide as a set of output pixel values to form an output feature map. S3.5: Then perform average pooling on the output feature map generated by the convolutional layer through the pooling layer to generate a reduced output feature map. Then, from shallow to deep, perform convolutional and pooling operations alternately multiple times until the final output feature map is output by the last pooling layer. The fully connected layer receives the output feature map flattened into a vector form, performs weighted fusion based on the preset weight matrix in the fully connected layer, and normalizes the fusion result into a probability distribution through the Softmax layer. S3.6: Calculate the total loss value of the lightweight quality detection model through the joint loss function based on the true quality labels and soft labels, and input the total loss value into the output layer of the lightweight quality detection model. Based on the chain rule, calculate the gradient values of the total loss value for each layer of the lightweight quality detection model through the backpropagation algorithm, and use the Adam optimizer to update the parameters of each layer of the lightweight quality detection model layer by layer. S3.7: After each round of periodic training, input the validation set into the trained lightweight quality detection model and evaluate the performance of the lightweight quality detection model. If the model performance does not reach the preset threshold, re-train using the training set and optimize the parameters of the lightweight quality detection model, and repeat the training and validation until the model performance reaches the preset threshold. Then, use the test set to test the trained lightweight quality detection model and deploy the lightweight quality detection model that meets the preset detection requirements to the corresponding edge computing device.

5. A system for monitoring the security management of tea quality data, used to implement the method for monitoring the security management of tea quality data according to any one of claims 1-4, characterized in that, Including: An acquisition and preprocessing module, a credibility evaluation module, a data fusion module, a data encryption module, an edge detection module, a detection and warning module, a collaborative optimization module, and a data archiving module; The acquisition and preprocessing module is used to acquire the quality indicators of environmental data, image data, and processing technology data, and preprocess the collected raw data; The credibility evaluation module dynamically calculates the credibility scores of the collected data according to the stability, historical performance, and real-time status of the preprocessed collected data; The data fusion module fuses the collected data according to the credibility scores of the collected data; The data encryption module is used to perform homomorphic encryption processing on the sensitive information in the collected data; The edge detection module is used to detect and identify tea images; The detection and warning module is used to detect situations such as abnormal fluctuations in quality data, sensor failures, and data offsets, and trigger multi-level warnings based on the detection results; The collaborative optimization module is used for joint training between multiple tea gardens and processing units, and collaboratively optimizes the parameters of the detection and warning module; The data archiving module generates summaries of each tea quality data and archives them on the blockchain to build a traceable path.

6. The tea quality data security management and monitoring system according to claim 5, characterized in that The specific steps for the data archiving module to build a traceable path are as follows: S4.1: Structurally process the tea quality data from different links, generate a unique identifier and a hash summary for each tea quality data, write the hash summaries at different stages into the planting chain, processing chain, and detection chain respectively, and control the operation of each chain through different central servers or the cloud; S4.2: When performing data comparison or traceability between different chains, initiate a verification request, use a relay contract or a light client for cross-chain communication, and at the same time build a zero-knowledge circuit between the chains to prove the consistency of the data on the chain. After the proof passes, only submit the zero-knowledge proofs of "verification successful" and "hash consistency"; S4.3: Another chain for data comparison or traceability receives the zero-knowledge proof of the cross-chain verification request and verifies it through a zero-knowledge verification engine. When the verification is successful, a verification record is automatically generated on the target chain, and the on-chain archiving and cross-chain verification results at each stage are collected in real time and integrated into a complete verifiable tea quality traceability path.

Images