An edible agricultural product traceability and supply chain management system

Abstract

The application discloses a kind of edible agricultural products traceability and supply chain management system, the present application relates to agricultural product supply chain informatization and data credible management technical field, including: data acquisition and authentication module, for from the multiple terminal equipment deployed in agricultural product production end Multimodal source data acquisition, and the identity authentication of production subject, the multimodal source data includes environmental sensing data, operation event data and key node image data.The edible agricultural products traceability and supply chain management system, the adaptability of model to diversified production scene and unknown abnormality is improved, realizes from production to the whole process coherent depiction and visual analysis of sale, via independent double engine verification and consensus mechanism processing data, cross-chain storage is authenticated after through oracle network, while guaranteeing sensitive information privacy, ensure the consistency and public credibility of traceability information in complex cooperation network.

Description

Technical Field

[0001] This invention relates to the field of agricultural product supply chain informatization and data trust management technology, specifically to a traceability and supply chain management system for edible agricultural products. Background Technology

[0002] As people's living standards continue to improve, consumers are paying increasing attention to food safety, especially the quality and safety of edible agricultural products. The quality of agricultural products has become one of the primary factors consumers consider when purchasing them. However, consumers and buyers often find it difficult to comprehensively, accurately, and effectively monitor the entire process of agricultural products from planting, harvesting, processing to distribution. Against this backdrop, the development of technologies such as digitalization, intelligentization, and big data has provided strong support for agricultural product traceability. More and more agricultural product supply chains are beginning to introduce digital technologies and build traceable supply chain information systems.

[0003] Several agricultural product traceability solutions already exist in the current technology. For example, the "Digital Traceability Method and System for Agricultural Product Supply Chain Information" document, published under the authorization number "CN119494669A," collects weather change information and harvesting and processing registration data. Based on an anomaly identification model, it analyzes the authenticity and rationality of the data and finally stores the processed data on the blockchain to achieve traceability and tamper-proofing of source data. This solution improves the transparency and credibility of the agricultural product supply chain to a certain extent, and has the advantages of rigorous data verification and a clear traceability chain.

[0004] However, the above solutions still have the following shortcomings: First, the system mainly relies on a pre-set anomaly data authentication model for data analysis. Model training depends on a large amount of labeled data. In actual application, if an uncovered planting pattern or environmental anomaly is encountered, the accuracy of identification may decrease. Second, the system does not adequately consider the user-friendliness of farmers. Data entry may still rely on manual operation, which poses a high barrier to entry for farmers with limited technical skills. Third, the system focuses on traceability in the production process and has weak integration and dynamic monitoring capabilities for subsequent links such as circulation, warehousing, and sales, making it difficult to achieve transparent management of the entire chain from farm to table. Fourth, although the system has the ability to upload data to the blockchain, it does not fully consider multi-chain collaboration and cross-chain data mutual recognition mechanisms, which may lead to data silos in complex supply chains involving multiple entities.

[0005] Therefore, there is an urgent need for a traceability and supply chain management system that can cover the entire life cycle of edible agricultural products, is easy to operate, and supports intelligent analysis and multi-chain collaboration, so as to further improve the quality and safety assurance capabilities of agricultural products and the overall operational efficiency of the supply chain. Summary of the Invention

[0006] The purpose of this invention is to provide a traceability and supply chain management system for edible agricultural products to solve the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a traceability and supply chain management system for edible agricultural products, comprising:

[0008] The data acquisition and authentication module is used to collect multimodal source data from multiple terminal devices deployed at the agricultural production end and to authenticate the identity of the production entity. The multimodal source data includes environmental sensor data, operation event data and key node image data.

[0009] An anomaly detection and learning module, which is communicatively connected to the data acquisition and authentication module, is used to aggregate local model parameters from the multiple terminal devices based on a federated learning framework to update the global anomaly detection model, and to send the updated model to the terminal devices for preliminary anomaly identification of the multimodal source data on the local terminal devices.

[0010] The full-chain monitoring and graph construction module is used to define and receive standardized event data reported by IoT devices in the warehousing, logistics and sales of agricultural products. Based on the multimodal source data and the standardized event data, it constructs and dynamically updates a traceability knowledge graph with agricultural product batches as entities and time-series events as edges.

[0011] The data verification and blockchain collaboration module is connected to the anomaly detection and learning module and the full-chain monitoring and graph construction module. It is used to perform asynchronous secondary verification on the preliminary anomaly identification results and the batch risk information output by the traceability knowledge graph, and write the verified data digest into multiple heterogeneous blockchain networks through the oracle network.

[0012] Furthermore, the terminal device integrates an environmental sensor, a sound acquisition unit, and an image acquisition unit;

[0013] The environmental sensor is used to continuously collect physical parameters of the crop growth environment;

[0014] The sound acquisition unit is used to capture the acoustic features corresponding to preset agricultural operations.

[0015] The image acquisition unit is used to automatically capture on-site images at key operational nodes;

[0016] The operation event data is generated by the production entity through the voice interface or preset operation button of the terminal device. The data acquisition and authentication module binds the environmental sensor data, successfully matched acoustic features, hash value of the on-site image and operation event data within the same time period to generate a multimodal evidence data package.

[0017] Furthermore, the anomaly detection and learning module is specifically used to perform the following operations:

[0018] On each of the terminal devices, the local anomaly detection sub-model is trained using the multimodal source data to generate model parameter updates;

[0019] The model parameters of each terminal device are updated and uploaded in encrypted form;

[0020] On the cloud aggregation server, model parameter updates uploaded by multiple terminal devices are aggregated to generate global model parameters;

[0021] The global model parameters are sent to the terminal device to update the local anomaly detection sub-model running locally.

[0022] The initial anomaly identification is performed offline on the terminal device by the updated local anomaly detection sub-model.

[0023] Furthermore, the full-chain monitoring and map construction module is specifically used for:

[0024] The definition covers various event types in the production, storage, logistics and sales of agricultural products, and specifies a standardized data format for each event type;

[0025] Receive event data that conforms to the standardized data format reported by warehouse environment sensors, logistics trajectory recorders and sales terminal data collection devices. The event data includes timestamps, spatial locations and business status information.

[0026] The multimodal source data is converted into event data for the production process;

[0027] The traceability knowledge graph is constructed by using agricultural product batch identifiers as graph nodes and event data from different stages as edges with time attributes.

[0028] Based on the aforementioned traceability knowledge graph, the state transitions and associated events of a specific batch of agricultural products throughout the entire supply chain are analyzed using a graph traversal algorithm.

[0029] Furthermore, the data verification and blockchain collaboration module includes a main verification engine, an independent audit engine, and a consensus contract;

[0030] The main verification engine performs the first verification based on the preliminary anomaly identification results output by the anomaly detection and learning module;

[0031] The independent audit engine performs a second verification based on the source knowledge graph output by the full-chain monitoring and graph construction module, applying algorithm rules different from those of the main verification engine.

[0032] The consensus contract is used to receive and compare the results of the first verification with those of the second verification.

[0033] If the comparison results are consistent, a final verification label is generated; if the comparison results are inconsistent, the main verification engine and the independent audit engine are triggered to recalculate based on the difference data and historical context until a consistent result is obtained or a manual audit process is triggered.

[0034] Furthermore, the data verification and blockchain collaboration module also includes an oracle network interface and a cross-chain relay;

[0035] The oracle network interface is used to submit the final verification label and key event data digest to a decentralized oracle network maintained by multiple trusted nodes for signature authentication.

[0036] The cross-chain relay is used to write data that has been signed and authenticated by the oracle network into the first blockchain corresponding to the production process, the second blockchain corresponding to the logistics process, and the third blockchain corresponding to the consumer query.

[0037] Furthermore, before writing the data to the blockchain, the data verification and blockchain collaboration module is also used to generate zero-knowledge proofs for data fields involving the privacy of the production entity, and to replace the original privacy data with the zero-knowledge proofs and write them to the blockchain through the cross-chain relay.

[0038] Furthermore, the system also includes a traceability information query service module, which is used to respond to query requests from consumer clients, obtain the full-chain event summary of the relevant agricultural product batch and the final verification label from the third blockchain, and display the key path of the traceability knowledge graph in the form of a graphical timeline.

[0039] Furthermore, the local anomaly detection sub-model of the terminal device is a pruned and optimized neural network model to adapt to the computing resources and power consumption limitations of the terminal device.

[0040] Furthermore, when constructing the traceability knowledge graph, the full-chain monitoring and graph construction module assigns different confidence weights to event data at different stages. These confidence weights are dynamically adjusted based on the reliability of the data source, the calibration status of the acquisition equipment, and the accuracy of historical data.

[0041] This invention provides a traceability and supply chain management system for edible agricultural products. It has the following beneficial effects:

[0042] This traceability and supply chain management system for edible agricultural products integrates multimodal data collection, federated learning, dynamic knowledge graphs, and blockchain collaboration mechanisms to construct a trusted traceability system covering the entire lifecycle of edible agricultural products. At the data source, it utilizes integrated terminals to automatically collect and correlate environmental, operational, and visual information. Through a federated learning framework, it enables the continuous collaborative evolution of decentralized anomaly detection models, effectively improving the models' adaptability to diverse production scenarios and unknown anomalies while protecting the data privacy of each production entity. Simultaneously, the system transforms information from each link in the supply chain into standardized events and constructs a dynamic knowledge graph, achieving a coherent depiction and visual analysis of the entire process from production to sales. Data processed through independent dual-engine verification and consensus mechanisms is stored cross-chain after authentication via an oracle network, ensuring the consistency and credibility of traceability information in a complex collaborative network while protecting the privacy of sensitive information.

[0043] This traceability and supply chain management system for edible agricultural products provides a complete, reliable, and verifiable data foundation for all participants in the agricultural product supply chain. Producers can contribute high-quality source data with low barriers to entry and benefit from precise production guidance. Regulators and distributors can obtain a transparent regulatory view and risk warnings throughout the entire process. Consumers can easily access intuitive and tamper-proof traceability information across the entire chain, enhancing consumer confidence. It improves the management efficiency and risk control level of agricultural product quality and safety, and also provides solid data support for establishing a mutually trusting supply chain collaborative ecosystem and derivative services. Attached Figure Description

[0044] Figure 1 This is a data flow diagram of a traceability and supply chain management system for edible agricultural products according to the present invention;

[0045] Figure 2 This is a module relationship diagram of a traceability and supply chain management system for edible agricultural products according to the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Please see Figure 1 and Figure 2 This invention provides a technical solution: a traceability and supply chain management system for edible agricultural products, comprising:

[0048] The data acquisition and authentication module is used to collect multimodal source data from multiple terminal devices deployed at the agricultural production end and to authenticate the identity of the production entity. The multimodal source data includes environmental sensor data, operational event data and key node image data.

[0049] The anomaly detection and learning module communicates with the data acquisition and authentication module. It is used to aggregate local model parameters from multiple terminal devices based on the federated learning framework to update the global anomaly detection model, and then distribute the updated model to the terminal devices for preliminary anomaly identification of multimodal source data on the local terminal devices.

[0050] The end-to-end monitoring and graph construction module is used to define and receive standardized event data reported by IoT devices in the warehousing, logistics and sales of agricultural products. Based on multimodal source data and standardized event data, it constructs and dynamically updates a traceability knowledge graph with agricultural product batches as entities and time-series events as edges.

[0051] The data verification and blockchain collaboration module, connected to the anomaly detection and learning module and the full-chain monitoring and graph construction module, is used to asynchronously verify the batch risk information output by the preliminary anomaly identification results and the traceability knowledge graph, and write the verified data summary into multiple heterogeneous blockchain networks through the oracle network.

[0052] It should be further explained that the system's data acquisition and authentication module obtains multimodal source data through terminal devices deployed at agricultural production sites. These terminal devices integrate environmental sensors, sound acquisition units, and image acquisition units. The environmental sensors continuously monitor and record physical parameters such as soil temperature and humidity, and light intensity. The sound acquisition units actively capture specific acoustic waveforms generated during harvesting, pruning, and other operations. The image acquisition units automatically capture on-site visual information at key operational nodes preset by the system or when manually triggered by the producer. At the same time, producers can mark operation events such as "start harvesting" and "irrigation completed" through simplified voice command interfaces or physical buttons on the terminal devices. This module performs spatiotemporal alignment and data binding on the environmental sensor readings, acoustic features identified by pattern matching, hash values ​​of on-site images, and manually marked operation events collected within the same time window, forming a multimodal evidence data package with inherent correlation and difficult to tamper with individually, and verifying the registered identity of the production entity and its declared planting area information.

[0053] The anomaly detection and learning module receives these data packets and operates using a federated learning architecture: each terminal device trains a miniaturized anomaly detection neural network model using locally stored multimodal evidence data packets, encrypting only the calculated model parameter gradients before uploading them to the cloud aggregation server; the aggregation server securely aggregates and averages the encrypted parameter gradients from multiple terminal devices to generate updated global model parameters, and then distributes these global parameters to each terminal device to update its local model; through continuous iteration, each terminal device can use its locally updated model to perform preliminary anomaly identification on real-time collected data, such as determining whether environmental readings deviate significantly from the reasonable range for crop growth, or whether the acoustic characteristics of the operation are inconsistent with the standard operating mode.

[0054] The end-to-end monitoring and graph construction module extends traceability from the production end to subsequent stages. This module predefines a series of standardized supply chain event types and their data formats, such as "temperature-controlled warehousing," "cold chain transportation," and "shelf display." It receives formatted event data proactively reported by IoT devices such as temperature and humidity recorders in warehouses, positioning and temperature sensors on transport vehicles, and smart shelves in sales locations. This data includes precise timestamps, device locations, and status readings. The module also converts multimodal evidence data packages from the production end into corresponding event nodes. Using a unique agricultural product batch identifier as the core, it dynamically constructs and maintains a visualized traceability knowledge graph by using event data generated in different stages and in chronological order as connection edges with directional and temporal attributes. Through this graph, all key event sequences of a batch of products from planting to sales can be traced.

[0055] The data verification and blockchain collaboration module is responsible for the final data trust processing and notarization. This module has a built-in main verification engine and an independent audit engine. The main verification engine produces a first conclusion based on the preliminary identification results of the anomaly detection and learning module, while the independent audit engine analyzes the knowledge graph topology and event logic relationships in the full-chain monitoring and graph construction module using different rule engines or graph algorithms to produce a second conclusion. A consensus contract deployed in the system compares these two conclusions. If they match, a final quality and risk label is generated. If they do not match, a recalculation process is initiated, and the difference data is fed back to the two engines for re-evaluation until a consensus is reached or manual intervention is required. For the final consensus conclusion and data summaries of key events, the system collects and jointly signs them for confirmation of rights through a decentralized oracle network composed of multiple pre-defined trusted institution nodes.

[0056] Subsequently, a cross-chain relay writes this trusted signed data into a first consortium chain serving production records, a second consortium chain serving logistics information, and a third public chain for public querying, according to business rules. For sensitive information involving the privacy of production entities, the system generates corresponding zero-knowledge proofs before writing them into the blockchain. Only the proofs are uploaded to the chain to ensure that the data is usable but the specific content is not visible, thereby realizing the secure, collaborative storage and verifiable flow of information across heterogeneous blockchains.

[0057] The terminal device integrates environmental sensors, a sound acquisition unit, and an image acquisition unit;

[0058] Environmental sensors are used to continuously collect physical parameters of the crop growth environment;

[0059] The sound acquisition unit is used to capture the acoustic features corresponding to preset agricultural operations;

[0060] The image acquisition unit is used to automatically capture on-site images at critical operational nodes;

[0061] Operation event data is generated by the production entity through the voice interface of the terminal device or preset operation buttons. The data acquisition and authentication module binds the environmental sensor data, successfully matched acoustic features, hash values ​​of on-site images and operation event data within the same time period to generate a multimodal evidence data package.

[0062] It should be further explained that the terminal device equipped in the data acquisition and authentication module is a dedicated hardware device integrating multiple sensing units. The environmental sensor module of this device continuously monitors and records physical parameters such as soil temperature, soil moisture, air temperature and humidity, and light intensity in the target farmland area. Its built-in sound acquisition unit, such as a high-sensitivity microphone, is responsible for actively acquiring audio signals in the working environment, and has a built-in pre-trained acoustic model for real-time matching and recognition of characteristic sound wave patterns corresponding to specific agricultural operations such as "shearing with scissors" and "fruit falling".

[0063] The image acquisition unit, typically a camera, automatically captures images or short videos of the work site at preset fixed times, triggered by sensor data (such as sudden changes in temperature and humidity), or manually triggered by the producer via a physical button on the device. When producers input data through this terminal device, there is no need to fill out complicated forms; they only need to press the physical button on the device panel labeled "Harvesting Start" or "Irrigation," or speak the corresponding preset voice command to generate a digital event record containing the operation type and timestamp.

[0064] The core processing unit of the data acquisition and authentication module assigns precise timestamps and geographic coordinates to each set of acquired data through its internal clock and positioning module. It associates and encapsulates environmental sensor data sequences from the same production batch within a preset time tolerance range, identified and confirmed acoustic feature markers of agricultural operations, encrypted hash values ​​of captured on-site images, and records of operation events triggered by the producer, generating a structured, multimodal evidence data package with a unified traceability identifier. The elements within this data package corroborate each other; any subsequent modification to a single piece of data will cause the internal association verification of the data package to fail, thus ensuring the integrity and tamper resistance of the source data acquisition.

[0065] The anomaly detection and learning module is specifically used to perform the following operations:

[0066] On each terminal device, the local anomaly detection sub-model is trained using multimodal source data to generate model parameter updates.

[0067] The model parameters of each terminal device are updated and uploaded in encrypted form;

[0068] On the cloud aggregation server, model parameter updates uploaded from multiple terminal devices are aggregated to generate global model parameters;

[0069] The global model parameters are sent to the terminal device to update the local anomaly detection sub-model running locally.

[0070] The initial anomaly identification is performed offline on the terminal device by the updated local anomaly detection sub-model.

[0071] It should be further explained that the anomaly detection and learning module implements a distributed model continuous optimization mechanism. On each production terminal device, an initialized lightweight anomaly detection model is stored and run locally. This model uses the multimodal evidence data packets generated by the data acquisition and authentication module and stored locally as the training sample set. Through forward computation and backpropagation algorithms, the model's parameter gradients or updates are calculated locally on the device. To protect data privacy and reduce communication burden, the terminal device does not upload the original multimodal evidence data packets. Instead, the calculated model parameter updates are encrypted and then uploaded to the aggregation server deployed in the cloud via the communication network.

[0072] The cloud-based aggregation server receives encrypted parameter updates uploaded from a large number of terminal devices within the region. Without decrypting the specific content, it uses a secure aggregation algorithm, such as weighted averaging of multiple update vectors, to calculate an aggregated global model parameter update. The aggregation server then uses this global update to generate a new version of the global anomaly detection model parameters.

[0073] Subsequently, the aggregation server distributes the updated global model parameters to all relevant terminal devices in the network. Upon receiving the new global model parameters, the terminal devices merge or replace them with their existing local model parameters, thereby completing the local model update. Through multiple rounds of this iterative process of "local training - encrypted upload - cloud aggregation - parameter distribution," the local model running on each terminal device can gradually absorb and learn common features and abnormal patterns from different production environments and operating modes. At the same time, the original data of each producer remains locally throughout the entire process, achieving collaborative model evolution under data privacy protection.

[0074] After the update is completed, the terminal device can use its latest local model to perform real-time analysis and preliminary anomaly identification on the newly collected multimodal source data, such as identifying combinations of environmental parameters that exceed the model's learning range or operation sequences that differ from normal operating modes.

[0075] The end-to-end monitoring and graph construction module is specifically used for:

[0076] The definition covers various event types in the production, storage, logistics and sales of agricultural products, and specifies a standardized data format for each event type;

[0077] Receive event data in a standardized format reported by warehouse environment sensors, logistics trajectory recorders and sales terminal data collection devices. The event data includes timestamps, spatial location and business status information.

[0078] Transform multimodal source data into event data for the production process;

[0079] Using agricultural product batch identifiers as graph nodes and event data from different stages as edges with time attributes, a traceability knowledge graph is constructed.

[0080] Based on the traceability knowledge graph, the state transition and related events of a specific batch of agricultural products in the whole chain are analyzed through graph traversal algorithm.

[0081] It should be further explained that the end-to-end monitoring and graph construction module realizes the event-based abstraction and correlation integration of data throughout the entire process. This module predefines several standardized event types covering key nodes in the agricultural product supply chain, such as "harvesting events at the production site," "warehouse entry events," "cold chain transportation departure events," "temperature-controlled storage events," and "retail shelf placement events," and specifies a unified structured data format for each event type. This format includes at least a unique event identifier, event occurrence time, geographical coordinates, operator identifier, associated agricultural product batch number, and a set of event-specific attributes.

[0082] This module continuously receives data messages from temperature and humidity sensor nodes deployed in the warehouse, electronic tags with positioning and temperature recording functions attached to transport packaging, and sensing devices on the smart shelves at the sales end through an adapter interface. These messages are encapsulated in the above-mentioned standardized format and reflect the status changes and operation actions of the corresponding links in real time or near real time.

[0083] Simultaneously, this module parses and transforms the multimodal evidence data packets generated by the data acquisition and authentication module into corresponding production process event data. The system uses each independent agricultural product batch identifier as a core entity node in the traceability knowledge graph, connecting various standardized event data occurring in a time sequence as edges. The direction of the edges indicates the event flow sequence, and the attributes on the edges record the detailed content of the event.

[0084] As new events are continuously added, the system dynamically adds new nodes and edges to the graph, or updates the state attributes of existing edges, thereby constructing a continuously growing dynamic graph that depicts the complete life cycle of agricultural products. Based on this graph, the system can use a pre-defined graph traversal algorithm to perform traceability path queries on a specified batch of agricultural products, clearly displaying the entire sequence of key events and their context from production to consumption. It can also use graph analysis techniques to identify specific patterns, such as detecting the temporal proximity between abnormal temperature events during transportation and subsequent storage events.

[0085] The data verification and blockchain collaboration module includes a main verification engine, an independent audit engine, and a consensus contract;

[0086] The main verification engine performs the first verification based on the preliminary anomaly identification results output by the anomaly detection and learning module;

[0087] The independent audit engine uses a traceability knowledge graph output by the full-chain monitoring and graph construction module to perform a second verification using algorithmic rules different from those of the main verification engine.

[0088] The consensus contract is used to receive and compare the results of the first verification with the second verification.

[0089] If the comparison results are consistent, a final verification label is generated; if the comparison results are inconsistent, the main verification engine and the independent audit engine are triggered to recalculate based on the difference data and historical context until a consistent result is obtained or a manual audit process is triggered.

[0090] It should be further explained that the data verification and blockchain collaboration module has a parallel dual-engine verification and consensus mechanism. The main verification engine receives the preliminary anomaly identification results from the anomaly detection and learning module. These results include anomaly identifiers and confidence levels derived from local analysis of multimodal data. The main verification engine then performs logical verification and formatting on these results according to a pre-defined rule base to generate the first verification conclusion.

[0091] The independent audit engine asynchronously receives real-time traceability knowledge graph data from the full-chain monitoring and graph construction module. Instead of directly using the original anomaly identification results, it analyzes the graph topology and attribute characteristics, such as the completeness of the event sequence of specific batch nodes, the spatiotemporal logical consistency between events, and the continuity of environmental parameters across events, and uses graph pattern matching or rule-based graph query language to generate independent second verification conclusions.

[0092] The consensus contract, as a pre-built programmatic component with comparison logic, receives conclusions from both the main verification engine and the independent audit engine. The contract's built-in comparison unit compares the two conclusions item by item across agreed-upon dimensions. If all comparison items fall within a preset consistency threshold, the consensus contract generates a unified, digitally signed final verification label. If the comparison unit detects a difference exceeding the threshold in either aspect, the consensus contract immediately triggers a recalculation process: it encapsulates the original data fragments associated with the difference and related historical context data into new computational task packages, which are then resent to both the main verification engine and the independent audit engine.

[0093] The two engines perform a new round of analysis and calculation based on this task package, and submit the new results to the consensus contract for comparison again. This cycle can continue until consensus is reached or the preset maximum number of cycles is reached. If consensus is not reached after the maximum number of cycles is reached, the consensus contract will generate a special event record containing details of all differences and trigger a manual audit request to be sent to the system's preset administrator interface, leaving the final decision to human intervention.

[0094] The data verification and blockchain collaboration module also includes an oracle network interface and a cross-chain relay;

[0095] The oracle network interface is used to submit the final verification label and key event data digest to a decentralized oracle network maintained by multiple trusted nodes for signature authentication;

[0096] Cross-chain relays are used to write data that has been signed and authenticated by the oracle network into the first blockchain corresponding to the production process, the second blockchain corresponding to the logistics process, and the third blockchain corresponding to the consumer query.

[0097] It should be further explained that the data verification and blockchain collaboration module further realizes data ownership confirmation and distributed notarization through oracle networks and cross-chain relay technology. The oracle network interface built into this module is responsible for encapsulating the final verification tag generated by the consensus contract and the key event data summary extracted from the full-chain monitoring and graph construction module into a data request to be confirmed, according to a preset format. This interface sends this request to a decentralized oracle network jointly maintained by multiple pre-selected and independent trusted institutional nodes. These nodes may include third-party quality inspection agencies, industry regulatory platforms, or designated certification centers.

[0098] Upon receiving a request, each node in the oracle network independently verifies the authenticity of the data in the request based on its own data source or verification rules. If the verification is successful, it digitally signs the data digest using its private key. When the number of valid signatures collected reaches a preset threshold, the oracle network packages these signatures with the original data digest to generate a trusted data packet with a federated authentication mark.

[0099] Subsequently, the cross-chain relay begins operation. Based on a pre-configured business rule mapping table, it routes and writes different parts of the aforementioned trusted data packets to different blockchain networks. For example, production-related data and their verification tags are written to a first consortium blockchain serving production records and supplier collaboration; logistics trajectory and temperature control event data are written to a second consortium blockchain serving mutual recognition among logistics companies; and product batch summary information and final quality labels for consumer inquiries are written to a transparent third public blockchain. The cross-chain relay interacts with node interfaces of each target blockchain network to complete transaction construction, transmission, and on-chain confirmation, thereby ensuring that the same traceability facts are synchronously and consistently recorded across multiple heterogeneous blockchain systems.

[0100] Before writing the data to the blockchain, the data verification and blockchain collaboration module is also used to generate zero-knowledge proofs for data fields involving the privacy of the production entity, and then write the zero-knowledge proofs to the blockchain in place of the original privacy data through the cross-chain relay.

[0101] It should be further explained that before writing data containing producer privacy information to the blockchain via the cross-chain relay, the data verification and blockchain collaboration module initiates a privacy protection process. The system has a pre-built privacy field identification rule base to automatically identify fields in the data to be uploaded to the blockchain that fall under the category of sensitive privacy, such as the producer's precise geographical coordinates, identity information, or specific business data. For each identified privacy field, the module calls an integrated zero-knowledge proof generation component, which is implemented based on a specific non-interactive zero-knowledge proof protocol. The generation process takes into account the original value of the privacy field, a publicly available set of verification parameters, and a publicly stated statement to be proven, such as "this geographical location is within a certified production area" or "the operator's identity has been verified through system registration." The proof generation component runs the corresponding cryptographic algorithm and outputs a fixed- or variable-length proof string that is independent of the length of the original data. This proof does not contain any valid information that can deduce the original privacy data.

[0102] Subsequently, during data encapsulation by the cross-chain relay, it completely replaces the corresponding privacy fields in the original data message with the generated zero-knowledge proof string. Finally, the cross-chain relay writes this replaced data structure, containing the proof instead of the original value, along with other non-privacy data, to the target blockchain. Any subsequent validators, including other on-chain participants or end consumers, can use the publicly available verification algorithm and the same publicly available verification parameters to verify this proof string attached to the chain, thereby confirming that the corresponding privacy data meets a specific authenticity condition, without ever touching the specific content of the privacy data.

[0103] The system also includes a traceability information query service module, which responds to query requests from consumer clients, obtains the full-chain event summary and final verification label of the relevant agricultural product batch from the third blockchain, and displays the key path of the traceability knowledge graph in the form of a graphical timeline.

[0104] It should be further explained that the traceability information query service module provides consumers with a transparent information verification portal. This module has a public-facing web service interface that receives query requests submitted by consumer clients. These requests typically include a unique traceability code for agricultural products obtained by scanning a QR code on the product packaging or by manually entering the code. Upon receiving the request, the service interface first parses the traceability code to extract the corresponding target batch identifier and associated blockchain address information.

[0105] Subsequently, the module invokes a node client connected to a third-party public blockchain network to query the specific smart contract status or transaction records stored on-chain related to the batch identifier, based on the obtained blockchain address information. The data packets obtained from the blockchain are parsed to reconstruct a "full-chain event summary" and a "final verification tag" authenticated by the oracle network. The event summary is a compressed representation of the critical path in the traceability knowledge graph, listing the main event nodes and their core attributes from production, processing, warehousing, transportation to sales in chronological order. The module has a built-in visualization rendering engine that automatically generates an interactive graphical timeline interface by combining the parsed event sequence and verification tags with predefined templates and icon libraries. In this interface, time is used as the horizontal axis, and different supply chain links are used as the vertical axis or category labels. Each key event is rendered as a node with an icon, and the flow order between nodes is indicated by lines. Clicking on a node expands to view the detailed attributes of that event.

[0106] The final verification label is displayed as a prominent badge at the top or side of the timeline. This graphical interface is returned to the consumer client via a web page or mobile application, presenting the complete and credible journey of agricultural products from source to end in an intuitive and structured way.

[0107] The local anomaly detection sub-model for terminal devices is a pruned and optimized neural network model to adapt to the computing resources and power consumption limitations of terminal devices.

[0108] It should be further explained that the anomaly detection sub-model running locally on the terminal equipment deployed at agricultural production sites is a specially designed and optimized lightweight neural network model, intended to match the device's own computing power, memory capacity, and battery life. The initial version of this model originated from a baseline model with strong generalization capabilities pre-trained in the cloud. Before deployment, it was processed using model compression techniques, such as channel pruning to remove low-contribution convolutional kernels or layer fusion to merge consecutive linear operation and activation function layers, to reduce the total number of parameters and computational complexity of the model.

[0109] Simultaneously, the weight parameters in the model may be quantized, for example, by representing the original 32-bit floating-point numbers as 8-bit integers, to further reduce the model's storage space and computational overhead during inference. After the above trimming and optimization, the generated lightweight model is packaged and embedded into the firmware or dedicated secure storage area of ​​the terminal device. When the terminal device starts up and runs, the model is loaded into the device's memory. After the data acquisition and authentication module generates a multimodal evidence data packet, the model is invoked to perform real-time calculations on the feature vectors or structured inputs in the data packet using a forward propagation method, outputting a judgment result on whether the data is normal or has potential anomalies and its corresponding confidence score. The lightweight nature of the model ensures that this local inference process can be completed quickly within the device's limited processor resources and energy budget, avoiding the latency and communication costs caused by continuously transmitting raw data to the cloud for parsing, while ensuring the availability of the system's core detection functions when network conditions are poor in the production site. The model's parameters can be iteratively optimized through global updates issued by the anomaly detection and learning module, but its basic architecture and lightweight characteristics remain unchanged during the update process, ensuring that it always adapts to the terminal's operating environment.

[0110] When constructing a traceability knowledge graph, the end-to-end monitoring and graph construction module assigns different confidence weights to event data at different stages. The confidence weights are dynamically adjusted based on the reliability of the data source, the calibration status of the acquisition equipment, and the accuracy of historical data.

[0111] It should be further explained that during the dynamic construction and updating of the traceability knowledge graph in the end-to-end monitoring and graph construction module, the system assigns a quantified confidence weight value to each piece of event data. This value is used to characterize the credibility and influence factor of the event data in subsequent graph analysis and reasoning. The initial assignment of the confidence weight is not fixed, but is dynamically generated and adjusted by an integrated weight calculation engine based on multi-dimensional real-time information. One core dimension is the reliability of the data source. The system maintains a list of trusted data sources and predefines different basic confidence coefficients for different types of collection devices (such as temperature and humidity sensors certified by authoritative institutions and ordinary consumer-grade sensors) and their affiliated institutions (such as data from official regulatory platforms and self-reported data from enterprises).

[0112] Another dimension is the real-time calibration status of the data acquisition devices. The system periodically or on-demand acquires the last calibration time, calibration certificate status, and self-test report of each IoT device through the device management interface. Based on this information, the system evaluates the device's current data output capability and maps it to a status coefficient. The third dimension is historical data accuracy. The system calculates the historical accuracy index of the data source by comparing data produced by the same device under similar conditions with verified benchmark data, or by analyzing the internal consistency and outlier ratio of its data in long-term sequences. The weight calculation engine periodically, or when new event data arrives, combines the coefficients of the above dimensions and calculates the final confidence weight of the event data according to a pre-set fusion algorithm (such as weighted average or rule-based minimum selection).

[0113] Subsequently, this weight value is attached as an attribute to the corresponding event node or edge. In subsequent graph queries, risk propagation calculations, or path analysis, the system will incorporate this confidence weight into the calculation when performing aggregation, inference, or sorting operations. For example, when calculating the comprehensive risk score of a batch, different contribution ratios are assigned to event data with different confidence levels, thereby making the graph-based analysis conclusions more reflective of the quality differences in the data itself and improving the robustness and accuracy of the entire source tracing judgment system.

[0114] This system integrates multimodal data acquisition, federated learning, dynamic knowledge graphs, and blockchain collaboration mechanisms to construct a trusted traceability system covering the entire lifecycle of edible agricultural products. At the data source, it utilizes integrated terminals to automatically collect and correlate environmental, operational, and visual information. Through a federated learning framework, it enables the continuous collaborative evolution of decentralized anomaly detection models, effectively improving the models' adaptability to diverse production scenarios and unknown anomalies while protecting the data privacy of each production entity. Simultaneously, the system transforms information from each link in the supply chain into standardized events and constructs a dynamic knowledge graph, achieving a coherent portrayal and visual analysis of the entire process from production to sales. Data processed through independent dual-engine verification and consensus mechanisms is stored cross-chain after authentication via an oracle network, ensuring the consistency and credibility of traceability information within a complex collaborative network while protecting the privacy of sensitive information.

[0115] The implementation of this system provides a complete, reliable, and verifiable data foundation for all participants in the agricultural product supply chain; producers can contribute high-quality source data in a low-barrier manner and benefit from precise production guidance; regulators and distributors can obtain a transparent regulatory view and risk warning throughout the entire process; consumers can easily obtain intuitive and tamper-proof traceability information across the entire chain, enhancing consumer confidence; it improves the management efficiency and risk control level of agricultural product quality and safety, and also provides solid data support for establishing a mutually trusting supply chain collaborative ecosystem and derivative services.

[0116] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 a process, method, article, or apparatus. Without further limitations, 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 said element.

[0117] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A traceability and supply chain management system for edible agricultural products, characterized in that, include: The data acquisition and authentication module is used to collect multimodal source data from multiple terminal devices deployed at the agricultural production end and to authenticate the identity of the production entity. The multimodal source data includes environmental sensor data, operation event data and key node image data. An anomaly detection and learning module, which is communicatively connected to the data acquisition and authentication module, is used to aggregate local model parameters from the multiple terminal devices based on a federated learning framework to update the global anomaly detection model, and to send the updated model to the terminal devices for preliminary anomaly identification of the multimodal source data on the local terminal devices. The full-chain monitoring and graph construction module is used to define and receive standardized event data reported by IoT devices in the warehousing, logistics and sales of agricultural products. Based on the multimodal source data and the standardized event data, it constructs and dynamically updates a traceability knowledge graph with agricultural product batches as entities and time-series events as edges. The data verification and blockchain collaboration module is connected to the anomaly detection and learning module and the full-chain monitoring and graph construction module. It is used to perform asynchronous secondary verification on the preliminary anomaly identification results and the batch risk information output by the traceability knowledge graph, and write the verified data digest into multiple heterogeneous blockchain networks through the oracle network.

2. The traceability and supply chain management system for edible agricultural products according to claim 1, characterized in that: The terminal device integrates an environmental sensor, a sound acquisition unit, and an image acquisition unit; The environmental sensor is used to continuously collect physical parameters of the crop growth environment; The sound acquisition unit is used to capture the acoustic features corresponding to preset agricultural operations. The image acquisition unit is used to automatically capture on-site images at key operational nodes; The operation event data is generated by the production entity through the voice interface or preset operation button of the terminal device. The data acquisition and authentication module binds the environmental sensor data, successfully matched acoustic features, hash value of the on-site image and operation event data within the same time period to generate a multimodal evidence data package.

3. The traceability and supply chain management system for edible agricultural products according to claim 1, characterized in that: The anomaly detection and learning module is specifically used to perform the following operations: On each of the terminal devices, the local anomaly detection sub-model is trained using the multimodal source data to generate model parameter updates; The model parameters of each terminal device are updated and uploaded in encrypted form; On the cloud aggregation server, model parameter updates uploaded by multiple terminal devices are aggregated to generate global model parameters; The global model parameters are sent to the terminal device to update the local anomaly detection sub-model running locally. The initial anomaly identification is performed offline on the terminal device by the updated local anomaly detection sub-model.

4. The traceability and supply chain management system for edible agricultural products according to claim 1, characterized in that: The full-chain monitoring and map construction module is specifically used for: The definition covers various event types in the production, storage, logistics and sales of agricultural products, and specifies a standardized data format for each event type; Receive event data that conforms to the standardized data format reported by warehouse environment sensors, logistics trajectory recorders and sales terminal data collection devices. The event data includes timestamps, spatial locations and business status information. The multimodal source data is converted into event data for the production process; The traceability knowledge graph is constructed by using agricultural product batch identifiers as graph nodes and event data from different stages as edges with time attributes. Based on the aforementioned traceability knowledge graph, the state transitions and associated events of a specific batch of agricultural products throughout the entire supply chain are analyzed using a graph traversal algorithm.

5. The traceability and supply chain management system for edible agricultural products according to claim 1, characterized in that: The data verification and blockchain collaboration module includes a main verification engine, an independent audit engine, and a consensus contract. The main verification engine performs the first verification based on the preliminary anomaly identification results output by the anomaly detection and learning module; The independent audit engine performs a second verification based on the source knowledge graph output by the full-chain monitoring and graph construction module, applying algorithm rules different from those of the main verification engine. The consensus contract is used to receive and compare the results of the first verification with those of the second verification. If the comparison results match, a final verification label is generated; If the comparison results are inconsistent, the main verification engine and the independent audit engine will be triggered to recalculate based on the difference data and historical context until a consistent result is obtained or a manual audit process is triggered.

6. The traceability and supply chain management system for edible agricultural products according to claim 5, characterized in that: The data verification and blockchain collaboration module also includes an oracle network interface and a cross-chain relay. The oracle network interface is used to submit the final verification label and key event data digest to a decentralized oracle network maintained by multiple trusted nodes for signature authentication. The cross-chain relay is used to write data that has been signed and authenticated by the oracle network into the first blockchain corresponding to the production process, the second blockchain corresponding to the logistics process, and the third blockchain corresponding to the consumer query.

7. The traceability and supply chain management system for edible agricultural products according to claim 6, characterized in that: Before writing the data to the blockchain, the data verification and blockchain collaboration module is also used to generate zero-knowledge proofs for data fields involving the privacy of the production entity, and to write the zero-knowledge proofs to the blockchain in place of the original privacy data through the cross-chain relay.

8. The traceability and supply chain management system for edible agricultural products according to claim 7, characterized in that: The system also includes a traceability information query service module, which is used to respond to query requests from consumer clients, obtain the full-chain event summary of the relevant agricultural product batch and the final verification label from the third blockchain, and display the key path of the traceability knowledge graph in the form of a graphical timeline.

9. The traceability and supply chain management system for edible agricultural products according to claim 1, characterized in that: The local anomaly detection sub-model of the terminal device is a pruned and optimized neural network model to adapt to the computing resources and power consumption limitations of the terminal device.

10. The traceability and supply chain management system for edible agricultural products according to claim 1, characterized in that: When constructing the traceability knowledge graph, the full-chain monitoring and graph construction module assigns different confidence weights to event data at different stages. The confidence weights are dynamically adjusted based on the reliability of the data source, the calibration status of the acquisition equipment, and the accuracy of historical data.

Images