Agricultural product deep processing quality traceability monitoring method and system based on big data

Abstract

This invention discloses a method and system for quality traceability monitoring of deep-processed agricultural products based on big data, belonging to the field of product processing traceability technology. It constructs a dynamically weighted deep-processing propagation identification model based on a spatiotemporal graph neural network, utilizes a graph attention mechanism to calculate the influence weights between nodes in real time, and constructs a knowledge-enhanced traceability reasoning model based on a GCN graph convolutional network. By collecting multimodal data including environmental parameters, equipment operating conditions, raw material batches, intermediate products, and finished product testing, the multimodal data is preprocessed to obtain standard multimodal data. A pre-defined prior knowledge graph of deep-processed agricultural products is fused with the real-time deep-processing propagation graph for reasoning to generate a quality processing report. The quality processing report is anchored to a blockchain storage platform in the form of a hash value, and a unique traceability code is generated and associated with the product packaging for consumers and regulators to scan and access. This enhances the regulatory transparency and public trust in deep-processed products.

Description

Technical Field

[0001] This invention relates to the field of product processing traceability technology, and in particular to a method and system for quality traceability and monitoring of deep-processed agricultural products based on big data. Background Technology

[0002] Deep processing of agricultural products is a crucial link in transforming primary agricultural products into high-value-added commodities. In this process, raw materials often come from multiple origins, are mixed in batches, and there are complex nonlinear coupling relationships between processing parameters. Traditional quality traceability systems mainly rely on barcodes or RFID to record static batch information for each process, using centralized databases for storage, supplemented by manual sampling. When the final product fails to meet quality standards, because raw materials from different batches in the deep processing chain are constantly merged and split, traditional batch-list-based traceability methods struggle to accurately pinpoint the root process and the batch of raw materials that contributed most to the anomaly. Existing data collection methods mostly involve sampling at discrete time points; if sensors malfunction or communication is interrupted, data loss occurs, leading to breaks in the subsequent traceability chain. Quality analysis results typically remain at the level of statistical reports, and production managers often have to rely on experience for trial and error, resulting in low efficiency in corrective action. Furthermore, consumers and regulators find it difficult to verify the authenticity and completeness of traceability information. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a method and system for quality traceability and monitoring of deep-processed agricultural products based on big data.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned method for quality traceability and monitoring of deep-processed agricultural products based on big data: Multimodal data, including environmental parameters, equipment operating conditions, raw material batches, intermediate products and finished product testing, are collected. The multimodal data is preprocessed to obtain standard multimodal data. A dynamic weighted deep processing propagation recognition model is constructed based on ST-GNN spatiotemporal graph neural network. The graph attention mechanism is used to calculate the influence weights between nodes in real time, and the evolution of node states in the time dimension is captured based on GRU gated recurrent units. The standard multimodal data is input into the deep processing propagation recognition model, and a real-time deep processing propagation map is output. A knowledge-enhanced tracing reasoning model is constructed based on GCN graph convolutional network. It integrates a pre-set prior knowledge graph of agricultural product deep processing with a real-time deep processing propagation graph to perform reasoning and generate a quality processing report. The quality processing report is anchored to the blockchain evidence storage platform in the form of a hash value, and a unique traceability code is generated and associated with the product packaging for consumers and regulators to scan and access.

[0005] Furthermore, in the aforementioned big data-based method for quality traceability and monitoring of deep-processed agricultural products, the method for constructing a dynamically weighted deep-processing propagation recognition model based on the ST-GNN spatiotemporal graph neural network, utilizing a graph attention mechanism to calculate the influence weights between nodes in real time, and capturing the evolution of node states over time based on GRU gated recurrent units, includes: The ST-GNN spatiotemporal graph neural network is modeled to abstract the deep processing flow into a dynamic graph structure, where nodes represent processing links, equipment and raw material batches, and edges represent the spatiotemporal relationships between them, including material flow, process dependence and time sequence. The system calculates the impact weight of each node on its neighboring nodes in real time and identifies key nodes through an attention mechanism, including the impact of equipment failure on downstream processes and the risk of a batch of raw material issues spreading to the quality of the finished product.

[0006] Furthermore, in the aforementioned big data-based method for quality traceability and monitoring of deep-processed agricultural products, the method for constructing a dynamically weighted deep-processing propagation recognition model based on the ST-GNN spatiotemporal graph neural network, utilizing a graph attention mechanism to calculate the influence weights between nodes in real time, and capturing the evolution of node states over time based on GRU gated recurrent units, includes: The GRU gated loop unit is used to capture the degree of change of each node's state over time. The node state includes at least temperature, humidity, and equipment load. The GRU gated loop unit controls the degree of influence of historical information through its reset and update gates, and dynamically adjusts the current state by combining the historical state and the current input using the GRU's gating mechanism.

[0007] Furthermore, in the aforementioned big data-based method for quality traceability and monitoring of deep-processed agricultural products, the collection of multimodal data includes environmental parameters, equipment operating conditions, raw material batches, intermediate products, and finished product testing. The multimodal data is then preprocessed to obtain standard multimodal data, including: By removing outliers, missing values, and duplicate data from the multimodal data, cleaned multimodal data is obtained. The data from different sources and formats in the cleaned multimodal data are uniformly converted into a standard format to obtain the converted multimodal data; Key features are extracted from the transformed multimodal data, including at least trend changes in environmental parameters, abnormal fluctuations in equipment operating conditions, and threshold exceedances in detection data, to obtain feature multimodal data; By integrating structured data, unstructured data, and time-series data from feature multimodal data, standard multimodal data is obtained.

[0008] Furthermore, in the aforementioned big data-based method for quality traceability and monitoring of deep-processed agricultural products, the step of inputting the standard multimodal data into the deep-processing propagation identification model and outputting a real-time deep-processing propagation map includes: Standard multimodal data is input into a dynamic weighted model, which combines spatiotemporal correlation and attention weights to calculate the propagation path and risk intensity between nodes in real time, generating a deep processing propagation map that intuitively displays the propagation path of quality problems in the processing chain, the affected nodes, and the risk level.

[0009] Furthermore, in the aforementioned big data-based method for quality traceability and monitoring of deep-processed agricultural products, the knowledge-enhanced traceability reasoning model constructed based on GCN graph convolutional networks integrates a pre-defined prior knowledge graph of deep-processed agricultural products with a real-time deep-processing propagation graph for reasoning, generating a quality processing report, including: A knowledge graph is constructed by integrating domain knowledge of deep processing of agricultural products. Domain knowledge includes at least process standards, regulatory requirements and historical quality cases. The knowledge graph contains entities, relationships and rules. By performing graph convolution operations on the real-time deep processing propagation graph and the prior knowledge graph, and by aggregating node information, the causes, scope of impact, and quality risk scores of potential quality problems are inferred, and a quality processing report is generated.

[0010] Furthermore, in the aforementioned big data-based method for quality traceability and monitoring of deep-processed agricultural products, the step of anchoring the quality processing report to a blockchain evidence storage platform in the form of a hash value, and simultaneously generating a unique traceability code and associating it with the product packaging for consumers and regulators to scan and access, includes: The quality processing report generates a unique digest value through a hash algorithm, which serves as an immutable digital fingerprint. This digest is then uploaded to a blockchain evidence storage platform, where the on-chain timestamp and verification node signature are recorded. A unique traceability code is generated based on the blockchain hash value and product information, and the traceability code is embedded into the product packaging. When consumers and regulators scan the product packaging, they can access the blockchain to verify the quality report in real time and trace the entire process information, including the source of raw materials, processing parameters, and testing records.

[0011] Furthermore, the agricultural product deep processing quality traceability and monitoring system based on big data includes the following modules: The data acquisition and processing module is used to collect multimodal data, including environmental parameters, equipment operating conditions, raw material batches, intermediate products and finished product testing, and to preprocess the multimodal data to obtain standard multimodal data. The recognition model building module is used to construct a dynamically weighted deep processing propagation recognition model based on the ST-GNN spatiotemporal graph neural network. It uses the graph attention mechanism to calculate the influence weights between nodes in real time and captures the evolution of node states in the time dimension based on the GRU gated recurrent unit. The propagation analysis and localization module is used to input the standard multimodal data into the deep processing propagation identification model and output a real-time deep processing propagation map; The knowledge-enhanced reasoning module is used to build a knowledge-enhanced tracing reasoning model based on the GCN graph convolutional network. It integrates the pre-set prior knowledge graph of agricultural product deep processing with the real-time deep processing propagation graph to perform reasoning and generate a quality processing report. The blockchain evidence storage and traceability module is used to anchor the quality processing report to the blockchain evidence storage platform in the form of a hash value, and at the same time generate a unique traceability code and associate it with the product packaging for consumers and regulators to scan and access.

[0012] Furthermore, in the big data-based agricultural product deep processing quality traceability and monitoring system, the knowledge-enhanced reasoning module includes the following sub-modules: The integration submodule is used to integrate domain knowledge of deep processing of agricultural products to build a knowledge graph. The domain knowledge includes at least process standards, regulatory requirements and historical quality cases. The knowledge graph contains entities, relationships and rules. The reasoning submodule is used to perform graph convolution operations on the real-time deep processing propagation graph and the prior knowledge graph. By aggregating node information, it infers the causes, scope of impact, and quality risk scores of potential quality problems, and generates a quality processing report.

[0013] Furthermore, in the big data-based agricultural product deep processing quality traceability and monitoring system, the blockchain-based evidence storage and traceability module includes the following sub-modules: The recording submodule is used to generate a unique digest value from the quality processing report using a hash algorithm, which serves as an immutable digital fingerprint. This digest is then uploaded to the blockchain evidence storage platform, recording the on-chain timestamp and the signature of the verification node. An encryption submodule is used to generate a unique traceability code based on the blockchain hash value and product information, and to embed the traceability code into the product packaging. The traceability submodule allows consumers and regulators to access blockchain-verified quality reports in real time after scanning the product packaging, tracing the entire process information including raw material sources, processing parameters, and testing records.

[0014] Its beneficial effects lie in the introduction of a quality anomaly propagation analysis algorithm based on spatiotemporal graph neural networks. This algorithm models each process unit in the deep processing chain as graph nodes and the material and energy transfer relationships as directed edges. It dynamically calculates the influence weights between nodes using a graph attention mechanism and captures temporal evolution patterns using gated loop units. This algorithm can automatically locate the root cause node and predict the propagation path of quality deviations at the first moment an anomaly occurs, solving the technical problem of difficulty in defining responsibility for anomalies caused by batch mixing and process coupling, and achieving a leap from post-event traceability to real-time deduction. A knowledge-enhanced tracing reasoning engine is constructed, integrating prior knowledge graphs in the field of agricultural product deep processing with real-time anomaly propagation graphs. Through joint reasoning using graph convolutional networks, it can generate interpretable reports containing root causes, the scope of affected batches, and corrective suggestions. Furthermore, the counterfactual reasoning module re-simulates the quality evolution process by masking or modifying node states one by one, and uses causal forests to evaluate the causal effect strength of each process parameter on the final quality non-compliance, thereby outputting a ranking of the most valuable corrective nodes for intervention. This provides a quantitative basis for production decisions and significantly reduces the trial-and-error costs caused by blind adjustments. A self-supervised contrastive learning algorithm is employed to augment and impute missing values ​​in the original multimodal data. A masked autoencoder is used to reconstruct obscured sensor data, maintaining data integrity even under sensor failure or communication interruption conditions, thus improving the system's robustness in real industrial environments. Anomaly reports and trusted end-to-end data are anchored to the blockchain in hash form and stored using a MerkleDAG data structure, supporting fine-grained parallel traceability for batch splitting and merging. Consumers can obtain tamper-proof and complete traceability information by scanning a unique traceability code, enhancing regulatory transparency and public trust. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 This is a schematic diagram of the first embodiment of the method for quality traceability and monitoring of deep-processed agricultural products based on big data in this invention. Figure 2 This is a schematic diagram of the second embodiment of the method for quality traceability and monitoring of deep-processed agricultural products based on big data in this invention. Figure 3 This is a schematic diagram of the first embodiment of the agricultural product deep processing quality traceability and monitoring system based on big data in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms "one," "an," and "this" used herein may also include the plural forms. It should be further understood that the terminology used in this specification includes the presence of features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, a big data-based method for quality traceability and monitoring of deep-processed agricultural products includes the following steps: Step 101: Construct a dynamically weighted deep processing propagation recognition model based on ST-GNN spatiotemporal graph neural network, use graph attention mechanism to calculate the influence weight between each node in real time, and capture the evolution law of node state in time dimension based on GRU gated recurrent unit. Specifically, in this embodiment, an ST-GNN spatiotemporal graph neural network is modeled to abstract the deep processing flow into a dynamic graph structure, where nodes represent processing links, equipment, and raw material batches, and edges represent the spatiotemporal relationships between them. The spatiotemporal relationships include material flow, process dependence, and time sequence. The influence weight of each node on its neighboring nodes is calculated in real time, and key nodes are identified through an attention mechanism, including the degree of impact of equipment failure on downstream processes and the risk of the spread of problems in a certain batch of raw materials to the quality of the finished product.

[0020] The GRU gated loop unit captures the degree of change of each node's state over time. The node state includes at least temperature, humidity, and equipment load. The GRU gated loop unit controls the degree of influence of historical information through reset and update gates, and uses the GRU's gating mechanism to dynamically adjust the current state in combination with historical state and current input.

[0021] A spatiotemporal graph neural network (ST-GNN) framework was completed, adapting to the complex production scenarios of agricultural product deep processing. All quality-influencing factors throughout the deep processing process are transformed into independent network nodes, including raw material batch nodes, processing equipment nodes for each step, workshop environmental monitoring nodes, semi-finished product process inspection nodes, finished product terminal quality inspection nodes, and manual operation station nodes. Combined with the actual factory production layout, process connection logic, and material flow paths, a complete spatial network topology is constructed, comprehensively covering all quality-related nodes throughout the deep processing process, building a basic network analysis framework that fits real-world production scenarios. A graph attention mechanism is introduced to achieve dynamic real-time calculation of node influence weights. The model captures the operating status, data fluctuations, and working condition changes of each network node in real time, autonomously learning and identifying the degree of correlation and influence between different nodes, abandoning the traditional static fixed weight analysis method. Based on the real-time production status, it can dynamically calculate the influence weight of each piece of equipment, each process, each environmental indicator, and each batch of raw materials on the quality of agricultural product processing, accurately distinguishing between direct and indirect influences, and primary and secondary influence nodes, updating the strength of correlations and transmission relationships between nodes in real time, and realistically reproducing the quality influence logic of various production factors in the processing process. Relying on gated loop units, GRUs capture the temporal evolution of node states. For the continuous dynamic data of each network node, it continuously tracks the dynamic evolution of node states with processing time, process progression, production rhythm adjustments, and environmental fluctuations. It records the dynamic changes in the data fluctuation amplitude, trend, and quality impact of each node across different production time periods, accurately capturing the temporal dynamic characteristics of the entire processing process. This achieves a deep integration of spatial node correlation analysis and temporal state evolution analysis, ultimately forming a deep processing propagation identification model with dynamic weighting and simultaneous spatiotemporal dual-dimensional analysis capabilities.

[0022] Step 102: Collect multimodal data including environmental parameters, equipment operating conditions, raw material batches, intermediate products and finished product testing, preprocess the multimodal data to obtain standard multimodal data, input the standard multimodal data into the deep processing propagation recognition model, and output the real-time deep processing propagation map. Specifically, in this embodiment, outliers, missing values, and duplicate data are removed from the multimodal data to obtain cleaned multimodal data; data from different sources and formats in the cleaned multimodal data are uniformly converted into a standard format to obtain transformed multimodal data; key features are extracted from the transformed multimodal data, including at least trend changes in environmental parameters, abnormal fluctuations in equipment operating conditions, and threshold exceedances in detection data, to obtain feature multimodal data; structured data, unstructured data, and time-series data in the feature multimodal data are integrated to obtain standard multimodal data.

[0023] Multiple types of data are collected from the entire process of deep processing of agricultural products. Environmental parameters, such as temperature, humidity, and light, are monitored in real time through sensors. Equipment operating conditions, including equipment running status, parameter settings, and maintenance records, are acquired by the industrial Internet of Things system. Raw material batches, including raw material source, batch number, and quality inspection report, are entered through the batch management system. Intermediate product and finished product testing data, including chemical composition, microbiological indicators, and sensory evaluation, are generated by laboratory testing equipment.

[0024] Remove outliers, missing values, and duplicate data to ensure data integrity. Convert data from different sources and formats to a standardized format, such as aligning timestamps, standardizing units, and conforming to coding conventions. Extract key features from the raw data, such as trend changes in environmental parameters, abnormal fluctuations in equipment operating conditions, and threshold exceedances in detection data. Integrate structured data, such as numerical parameters, and unstructured data, such as text records, images, and time-series data, into a unified standard multimodal dataset to facilitate subsequent model processing.

[0025] Standard multimodal data is input into a dynamic weighted model, which combines spatiotemporal correlation and attention weights to calculate the propagation path and risk intensity between nodes in real time, generating a deep processing propagation map that intuitively displays the propagation path of quality problems in the processing chain, the affected nodes, and the risk level.

[0026] Preprocessed standard multimodal data is input into a dynamic weighted model. The model combines spatiotemporal correlation and attention weights to calculate the propagation path and risk intensity between nodes in real time. A dynamic, in-depth processing propagation map is generated, which intuitively displays the propagation path, affected nodes, and risk levels of quality problems such as contamination and parameter anomalies in the processing chain, helping to quickly locate the source of the problem.

[0027] Step 103: Construct a knowledge-enhanced tracing reasoning model based on GCN graph convolutional network, integrate the pre-set prior knowledge graph of agricultural product deep processing with the real-time deep processing propagation graph for reasoning, and generate a quality processing report; Specifically, in this embodiment, domain knowledge of deep processing of agricultural products is integrated to construct a knowledge graph. The domain knowledge includes at least process standards, regulatory requirements and historical quality cases. The knowledge graph contains entities, relationships and rules. The real-time deep processing propagation graph is convolved with the prior knowledge graph. By aggregating node information, the causes, scope of impact and quality risk scores of potential quality problems are inferred, and a quality processing report is generated.

[0028] We establish the basic architecture for a traceability reasoning model based on Graph Convolutional Networks (GCNs). Leveraging the powerful graph data feature extraction and correlation reasoning capabilities of GCNs, we adapt the model to the traceability scenario of agricultural product deep processing quality. We specifically adapt and optimize the model for node features, link relationships, and spatiotemporal data features of the deep processing propagation graph, and construct a basic reasoning model that can realize correlation reasoning, anomaly tracing, and causal analysis.

[0029] A comprehensive prior knowledge graph for deep processing of agricultural products is constructed, covering industry standards, process specifications, causes of common quality problems, thresholds of factor influence, corresponding processes for quality defects, and risk levels of abnormal operating conditions for various agricultural product processing. This knowledge graph is deeply integrated with the real-time deep processing propagation graph generated by the model, allowing the model to perform dual reasoning by combining industry expertise with real-time production data. This avoids the limitations of pure data computation and improves the professionalism and accuracy of traceability reasoning.

[0030] By employing graph convolutional networks to perform deep feature mining and correlation reasoning on fused data, the core influencing nodes of product quality, abnormal processes, causes of problems, and risk propagation paths are accurately located. Simultaneously, the compliance, process stability, and quality controllability of each production stage are analyzed. Based on all reasoning results, a standardized and detailed batch quality processing report is automatically generated, comprehensively recording the batch production overview, quality data at each stage, abnormal issues, traceability conclusions, quality rating, and optimization suggestions.

[0031] Step 104: Anchor the quality processing report to the blockchain evidence storage platform in the form of a hash value, and at the same time generate a unique traceability code and associate it with the product packaging for consumers and regulators to scan and access.

[0032] Specifically, in this embodiment, the quality processing report is used to generate a unique digest value through a hash algorithm, which serves as an immutable digital fingerprint. This digest is then uploaded to a blockchain evidence storage platform, where the on-chain timestamp and verification node signature are recorded. A unique traceability code is generated based on the blockchain hash value and product information, and this code is embedded in the product packaging. When consumers and regulators scan the code on the product packaging, they can access the blockchain to verify the quality report in real time, tracing the entire process information of raw material sources, processing parameters, and testing records.

[0033] The generated batch quality processing reports and full-process traceability data are encrypted and processed to generate a unique hash value. This hash value corresponds one-to-one with the batch data and report content. If the data is tampered with, the hash value will change significantly. This hash value is then uploaded and anchored to a blockchain evidence storage platform in real time to permanently store the batch processing quality data and traceability reports. Relying on the decentralized, immutable, and fully traceable characteristics of blockchain, all traceability data is guaranteed to be authentic, verifiable, and tamper-proof.

[0034] For each batch of processed agricultural products, the system generates a unique, non-repeatable, and lifelong valid traceability QR code based on blockchain-stored information and batch-specific production data. This traceability code is then precisely linked to the corresponding batch's blockchain-stored data, quality processing report, and full-process traceability data. Simultaneously, the unique traceability code is printed and affixed to the outer packaging of the final product for the corresponding batch, achieving one code per item and one certificate per batch for traceability identification management.

[0035] Consumers can easily scan a code with their mobile phones to access the full-process traceability information and quality report of the product, from the origin of raw materials, processing environment, equipment conditions, process flow, quality inspection results to quality rating; market supervision departments can verify the authenticity of product data and query blockchain-stored information through the traceability code to quickly carry out product quality supervision, problem batch verification, and quality responsibility determination.

[0036] Its beneficial effects lie in the introduction of a quality anomaly propagation analysis algorithm based on spatiotemporal graph neural networks. This algorithm models each process unit in the deep processing chain as graph nodes and the material and energy transfer relationships as directed edges. It dynamically calculates the influence weights between nodes using a graph attention mechanism and captures temporal evolution patterns using gated loop units. This algorithm can automatically locate the root cause node and predict the propagation path of quality deviations at the first moment an anomaly occurs, solving the technical problem of difficulty in defining responsibility for anomalies caused by batch mixing and process coupling, and achieving a leap from post-event traceability to real-time deduction. A knowledge-enhanced tracing reasoning engine is constructed, integrating prior knowledge graphs in the field of agricultural product deep processing with real-time anomaly propagation graphs. Through joint reasoning using graph convolutional networks, it can generate interpretable reports containing root causes, the scope of affected batches, and corrective suggestions. Furthermore, the counterfactual reasoning module re-simulates the quality evolution process by masking or modifying node states one by one, and uses causal forests to evaluate the causal effect strength of each process parameter on the final quality non-compliance, thereby outputting a ranking of the most valuable corrective nodes for intervention. This provides a quantitative basis for production decisions and significantly reduces the trial-and-error costs caused by blind adjustments. A self-supervised contrastive learning algorithm is employed to augment and impute missing values ​​in the original multimodal data. A masked autoencoder is used to reconstruct obscured sensor data, maintaining data integrity even under sensor failure or communication interruption conditions, thus improving the system's robustness in real industrial environments. Anomaly reports and trusted end-to-end data are anchored to the blockchain in hash form and stored using a MerkleDAG data structure, supporting fine-grained parallel traceability for batch splitting and merging. Consumers can obtain tamper-proof and complete traceability information by scanning a unique traceability code, enhancing regulatory transparency and public trust.

[0037] Please see Figure 2 In the big data-based method for quality traceability and monitoring of deep-processed agricultural products, multimodal data including environmental parameters, equipment operating conditions, raw material batches, intermediate products, and finished product testing are collected. Preprocessing of the multimodal data to obtain standard multimodal data includes the following steps: Step 201: Remove outliers, missing values, and duplicate data from the multimodal data to obtain cleaned multimodal data; Step 202: Convert the data from different sources and formats in the cleaned multimodal data into a standard format to obtain the converted multimodal data; Step 203: Extract key features from the transformed multimodal data, including at least the trend changes of environmental parameters, abnormal fluctuations in equipment operating conditions, and threshold exceedances of detection data, to obtain feature multimodal data; Step 204: Integrate the structured data, unstructured data, and time-series data in the feature multimodal data to obtain standard multimodal data.

[0038] The above describes embodiments of the big data-based quality traceability and monitoring method for deep processing of agricultural products according to the present invention. Please refer to [link / reference]. Figure 3 In the big data-based quality traceability and monitoring system for deep-processed agricultural products, the system includes the following modules: The data acquisition and processing module is used to collect multimodal data, including environmental parameters, equipment operating conditions, raw material batches, intermediate products and finished product testing, and to preprocess the multimodal data to obtain standard multimodal data. The recognition model building module is used to construct a dynamically weighted deep processing propagation recognition model based on the ST-GNN spatiotemporal graph neural network. It uses the graph attention mechanism to calculate the influence weights between nodes in real time and captures the evolution of node states in the time dimension based on the GRU gated recurrent unit. The propagation analysis and localization module is used to input standard multimodal data into the deep processing propagation identification model and output a real-time deep processing propagation map; The knowledge-enhanced reasoning module is used to build a knowledge-enhanced tracing reasoning model based on the GCN graph convolutional network. It integrates the pre-set prior knowledge graph of agricultural product deep processing with the real-time deep processing propagation graph to perform reasoning and generate a quality processing report. The blockchain evidence storage and traceability module is used to anchor the quality processing report to the blockchain evidence storage platform in the form of a hash value, and at the same time generate a unique traceability code and associate it with the product packaging for consumers and regulators to scan and access.

[0039] Specifically, this embodiment also achieves this through the following steps: Step S1: Construct a multi-source heterogeneous data acquisition layer. At the key control points of the entire process of agricultural product deep processing, the CCP deploys IoT sensors and edge computing nodes to collect multi-modal data in real time, including environmental parameters, equipment operating conditions, raw material batches, intermediate products and finished product detection. The multi-modal data is then timestamped and semantically labeled to form the original traceability data stream. Step S2 introduces a quality anomaly propagation analysis algorithm based on spatiotemporal graph neural network (ST-GNN). Each process unit in the deep processing chain is defined as a graph node, and the relationship between material flow and energy transfer is defined as an edge, constructing a dynamic weighted directed graph model. The graph attention mechanism is used to calculate the influence weight between each node in real time, and based on the gated recurrent unit (GRU), the evolution law of node state in the time dimension is captured, so as to automatically locate the root cause node of the anomaly and predict the propagation path of quality deviation when an anomaly occurs. Step S3: Based on the root cause nodes of the anomalies located in step S2, a knowledge-enhanced tracing reasoning engine is constructed. This engine integrates a pre-set prior knowledge graph of agricultural product deep processing, including raw material contraindications, process standards, microbial risk models, and the real-time anomaly propagation graph output in step S2. It uses graph convolutional networks for joint reasoning to generate an interpretable quality anomaly report. The report includes the root cause of the anomaly, the range of affected product batches, and recommended corrective measures. Step S4: The anomaly report and corresponding trusted data throughout the entire process generated in step S3 are anchored to the blockchain evidence storage platform in the form of hash values. At the same time, a unique traceability code is generated and associated with the final product packaging, allowing consumers or regulators to access the blockchain evidence storage platform by scanning the code to obtain tamper-proof traceability information and quality evaluation results.

[0040] Step S1 also includes using a self-supervised contrastive learning algorithm to perform data augmentation and missing value imputation on the original source data stream; specifically, using a masked autoencoder (MAE) to randomly mask part of the sensor time-series data, and reconstructing the masked data through an encoder-decoder structure, so as to learn the intrinsic representation of the data under unlabeled conditions, thereby improving the robustness of the data acquisition layer in the event of sensor failure or communication interruption.

[0041] The spatiotemporal graph neural network, ST-GNN, in step S2 employs a loss function jointly optimized by dynamic time warping, DTW, and graph edit distance during its training process. This function measures the similarity of graph structure changes during different batch processing, enabling the model to adaptively identify process graph structure drift caused by different raw material sources and different operators, and dynamically adjust the graph attention parameters accordingly.

[0042] The knowledge-enhanced tracing reasoning engine in step S3 also includes a counterfactual reasoning module. When the final product quality is found to be substandard, this module uses counterfactual intervention to block or modify certain node states of the graph model in step S2 one by one, re-simulates the quality evolution process, and uses causal forest to evaluate the causal effect strength of each node state on the final quality substandard, thereby outputting the ranking of the most valuable corrective nodes.

[0043] In step S4, the data structure anchored to the blockchain is a Merkle DAG, a directed acyclic graph structure, to support fine-grained batch splitting and merging records. When raw materials from multiple upstream batches are mixed and enter the same production tank, the hash pointers of the Merkle DAG point to the historical blocks of each raw material batch, so that a single query for the final product can trace the complete historical path of all related raw materials in parallel.

[0044] After processing and final quality inspection of N batches of products, the spatiotemporal graph neural network in step S2 and the graph convolutional network in step S3 are incrementally fine-tuned using the newly generated real quality data. During the fine-tuning process, the EWC algorithm is reinforced with elastic weights to limit large changes in key network parameters and prevent catastrophic forgetting of the original process knowledge.

[0045] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and equivalents.

Claims

1. A method for quality traceability and monitoring of deep-processed agricultural products based on big data, characterized in that, The method for quality traceability and monitoring of deep-processed agricultural products includes the following steps: A dynamic weighted deep processing propagation recognition model is constructed based on ST-GNN spatiotemporal graph neural network. The graph attention mechanism is used to calculate the influence weights between nodes in real time, and the evolution of node states in the time dimension is captured based on GRU gated recurrent units. Multimodal data, including environmental parameters, equipment operating conditions, raw material batches, intermediate products, and finished product testing, is collected. The multimodal data is preprocessed to obtain standard multimodal data. The standard multimodal data is then input into the deep processing propagation recognition model to output a real-time deep processing propagation map. A knowledge-enhanced tracing reasoning model is constructed based on GCN graph convolutional network. It integrates a pre-set prior knowledge graph of agricultural product deep processing with a real-time deep processing propagation graph to perform reasoning and generate a quality processing report. The quality processing report is anchored to the blockchain evidence storage platform in the form of a hash value, and a unique traceability code is generated and associated with the product packaging for consumers and regulators to scan and access.

2. The method for quality traceability and monitoring of deep-processed agricultural products based on big data as described in claim 1, characterized in that, The aforementioned ST-GNN spatiotemporal graph neural network-based dynamically weighted deep processing propagation recognition model utilizes a graph attention mechanism to calculate the influence weights between nodes in real time, and captures the evolution of node states over time based on GRU gated recurrent units, including: The ST-GNN spatiotemporal graph neural network is modeled to abstract the deep processing flow into a dynamic graph structure, where nodes represent processing links, equipment and raw material batches, and edges represent the spatiotemporal relationships between them, including material flow, process dependence and time sequence. The system calculates the impact weight of each node on its neighboring nodes in real time and identifies key nodes through an attention mechanism, including the impact of equipment failure on downstream processes and the risk of a batch of raw material issues spreading to the quality of the finished product.

3. The method for quality traceability and monitoring of deep-processed agricultural products based on big data as described in claim 1, characterized in that, The aforementioned ST-GNN spatiotemporal graph neural network-based dynamically weighted deep processing propagation recognition model utilizes a graph attention mechanism to calculate the influence weights between nodes in real time, and captures the evolution of node states over time based on GRU gated recurrent units, including: The GRU gated loop unit is used to capture the degree of change of each node's state over time. The node state includes at least temperature, humidity, and equipment load. The GRU gated loop unit controls the degree of influence of historical information through its reset and update gates, and dynamically adjusts the current state by combining the historical state and the current input using the GRU's gating mechanism.

4. The method for quality traceability and monitoring of deep-processed agricultural products based on big data as described in claim 1, characterized in that, The data collection includes multimodal data such as environmental parameters, equipment operating conditions, raw material batches, intermediate products, and finished product testing. The multimodal data is preprocessed to obtain standard multimodal data, including: By removing outliers, missing values, and duplicate data from the multimodal data, cleaned multimodal data is obtained. The data from different sources and formats in the cleaned multimodal data are uniformly converted into a standard format to obtain the converted multimodal data; Key features are extracted from the transformed multimodal data, including at least trend changes in environmental parameters, abnormal fluctuations in equipment operating conditions, and threshold exceedances in detection data, to obtain feature multimodal data. By integrating structured data, unstructured data, and time-series data from feature multimodal data, standard multimodal data is obtained.

5. The method for quality traceability and monitoring of deep-processed agricultural products based on big data as described in claim 1, characterized in that, The step of inputting the standard multimodal data into the deep processing propagation recognition model and outputting a real-time deep processing propagation map includes: Standard multimodal data is input into a dynamic weighted model, which combines spatiotemporal correlation and attention weights to calculate the propagation path and risk intensity between nodes in real time, generating a deep processing propagation map that intuitively displays the propagation path of quality problems in the processing chain, the affected nodes, and the risk level.

6. The method for quality traceability and monitoring of deep-processed agricultural products based on big data as described in claim 1, characterized in that, The knowledge-enhanced tracing reasoning model based on GCN graph convolutional network integrates a pre-defined prior knowledge graph of agricultural product deep processing with a real-time deep processing propagation graph for reasoning, generating a quality processing report, including: A knowledge graph is constructed by integrating domain knowledge of deep processing of agricultural products. Domain knowledge includes at least process standards, regulatory requirements and historical quality cases. The knowledge graph contains entities, relationships and rules. By performing graph convolution operations on the real-time deep processing propagation graph and the prior knowledge graph, and by aggregating node information, the causes, scope of impact, and quality risk scores of potential quality problems are inferred, and a quality processing report is generated.

7. The method for quality traceability and monitoring of deep-processed agricultural products based on big data as described in claim 1, characterized in that, The process of anchoring the quality processing report to a blockchain evidence storage platform in the form of a hash value, and simultaneously generating a unique traceability code and associating it with the product packaging for consumers and regulators to scan and access, includes: The quality processing report generates a unique digest value through a hash algorithm, which serves as an immutable digital fingerprint. This digest is then uploaded to a blockchain evidence storage platform, where the on-chain timestamp and verification node signature are recorded. A unique traceability code is generated based on the blockchain hash value and product information, and the traceability code is embedded into the product packaging. When consumers and regulators scan the product packaging, they can access the blockchain to verify the quality report in real time and trace the entire process information, including the source of raw materials, processing parameters, and testing records.

8. A big data-based quality traceability and monitoring system for deep-processed agricultural products, characterized in that: The agricultural product deep processing quality traceability and monitoring system includes the following modules: The recognition model building module is used to construct a dynamically weighted deep processing propagation recognition model based on the ST-GNN spatiotemporal graph neural network. It uses the graph attention mechanism to calculate the influence weights between nodes in real time and captures the evolution of node states in the time dimension based on the GRU gated recurrent unit. The propagation analysis and positioning module is used to collect multimodal data including environmental parameters, equipment operating conditions, raw material batches, intermediate products and finished product detection, preprocess the multimodal data to obtain standard multimodal data, input the standard multimodal data into the deep processing propagation identification model, and output a real-time deep processing propagation map. The knowledge-enhanced reasoning module is used to build a knowledge-enhanced tracing reasoning model based on the GCN graph convolutional network. It integrates the pre-set prior knowledge graph of agricultural product deep processing with the real-time deep processing propagation graph to perform reasoning and generate a quality processing report. The blockchain evidence storage and traceability module is used to anchor the quality processing report to the blockchain evidence storage platform in the form of a hash value, and at the same time generate a unique traceability code and associate it with the product packaging for consumers and regulators to scan and access.

9. The agricultural product deep processing quality traceability and monitoring system based on big data as described in claim 8, characterized in that, The knowledge-enhanced reasoning module includes the following sub-modules: The integration submodule is used to integrate domain knowledge of deep processing of agricultural products to build a knowledge graph. The domain knowledge includes at least process standards, regulatory requirements and historical quality cases. The knowledge graph contains entities, relationships and rules. The reasoning submodule is used to perform graph convolution operations on the real-time deep processing propagation graph and the prior knowledge graph. By aggregating node information, it infers the causes, scope of impact, and quality risk scores of potential quality problems, and generates a quality processing report.

10. The agricultural product deep processing quality traceability and monitoring system based on big data as described in claim 8, characterized in that, The blockchain-based evidence storage and traceability module includes the following sub-modules: The recording submodule is used to generate a unique digest value from the quality processing report using a hash algorithm, which serves as an immutable digital fingerprint. This digest is then uploaded to the blockchain evidence storage platform, recording the on-chain timestamp and the signature of the verification node. An encryption submodule is used to generate a unique traceability code based on the blockchain hash value and product information, and to embed the traceability code into the product packaging. The traceability submodule allows consumers and regulators to access blockchain-verified quality reports in real time after scanning the product packaging, tracing the entire process information including raw material sources, processing parameters, and testing records.

Images