Agricultural product quality risk early warning method and system based on multi-source data fusion
By constructing multi-source datasets and machine learning models, the limitations of existing agricultural product quality testing methods have been overcome, enabling risk assessment of multiple contaminants across the entire agricultural product chain, thereby improving food safety and resource utilization efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-31
- Publication Date
- 2026-03-10
AI Technical Summary
Existing agricultural product quality testing methods rely solely on pesticide residues, failing to effectively assess contamination from other sources such as chemical pollutants, heavy metals, mycotoxins, nitrites, antibiotics, hormones, pathogenic microorganisms, and parasite eggs, making it difficult to guarantee food safety.
We construct multi-source heterogeneous datasets, monitor soil and agricultural product data across the entire supply chain using IoT sensors and portable devices, apply machine learning algorithms to train risk assessment models, predict the probability of various pollution risks, and connect with market mechanisms for dynamic sampling and risk management.
It enables precise assessment of the risks of multiple pollutants across the entire agricultural product chain, improves food safety, optimizes sampling strategies, incentivizes producers to improve quality, reduces resource waste, and achieves environmental benefits through a carbon credit mechanism.
Smart Images
Figure CN121638862A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the field of quality detection, and particularly relates to a method and system for early warning of agricultural product quality risks based on multi-source data fusion. BACKGROUND
[0002] The existing agricultural products are often judged by detecting pesticide residues to determine whether the quality of the agricultural products is qualified, but the object of pesticide residue detection is artificially applied pesticides. However, the pollutants on the agricultural products are far more than that.
[0003] 1. Chemical pollutants other than pesticides
[0004] Heavy metal pollution: such as lead, cadmium, mercury, arsenic, etc. These mainly come from contaminated soil and water sources (such as industrial wastewater, mine wastewater).
[0005] Mycotoxins: such as aflatoxins (common in peanuts, corn), ochratoxins, etc. These are natural toxins produced by mold during the storage of grains and nuts, are strong carcinogens, and cannot be completely removed by conventional cleaning and cooking.
[0006] Nitrite: especially for leafy vegetables, excessive use of nitrogen fertilizer will lead to accumulation of nitrite, which has potential risks to human health.
[0007] Antibiotics and hormones: mainly used in livestock and aquaculture to prevent animal diseases or promote growth. Their residues can enter the human body through meat, eggs, and milk, leading to drug resistance and endocrine disorders.
[0008] 2. Biological pollution
[0009] Pathogenic microorganisms: such as E. coli, Salmonella, Listeria, etc. These contaminations often occur during post-harvest cleaning, processing, and transportation (such as using contaminated water for cleaning, and substandard hygiene of operators). Consumption can cause acute food poisoning. Fresh salad vegetables, pre-cut fruits, etc. are at high risk.
[0010] Parasite eggs: such as vegetables for raw consumption (lettuce, coriander, etc.) may carry roundworm eggs, liver fluke eggs, etc. if fertilized with human and animal manure or contaminated water.
[0011] Therefore, there are obvious limitations in using pesticide residue detection as the only criterion for judging the quality of agricultural products, which is difficult to ensure food safety, and needs to be improved. SUMMARY
[0012] Therefore, it is necessary to provide a method and system for early warning of agricultural product quality risks based on multi-source data fusion in view of the above problems.
[0013] The embodiment of the present application is implemented in a kind of agricultural product quality risk early warning method based on multi-source data fusion, comprising the following steps:
[0014] Multi-source heterogeneous data set is constructed, in prenatal link, (can utilize the Internet of Things sensor deployed in field to monitor continuously, also can use low cost, portable soil heavy metal quick tester, pH test paper etc. to carry out regular inspection) first data in soil is monitored, first data includes heavy metal (such as cadmium, lead, mercury, arsenic) content, pH value;In production link, second data is automatically recorded, second data includes water quality report of irrigation water source, application type and time of chemical fertilizer and pesticide;In postnatal link, third data is collected, third data includes harvesting time, temperature and humidity variation of storage warehouse, environmental conditions in transportation process;Microbial monitoring is carried out to water source and personnel health possibly contacted in all links, and fourth data is obtained;
[0015] Multi-source heterogeneous data is cleaned, aligned and associated, is converted into standardization data stream that can be used for analysis, through unified space mark (such as plot code) and time stamp (such as batch number) as index, the first data, second data, third data, fourth data of the same batch agricultural product whole chain are matched, and a fusion database with agricultural product batch as unique mark is constructed;
[0016] Based on fusion database, machine learning algorithm is applied to train risk assessment model, and the risk probability of each batch agricultural product facing various pollutions (such as heavy metal enrichment, fungal toxin breeding, microbial overproof) is predicted through risk assessment model, and the risk probability and risk type output by risk assessment model are used as the core basis for dynamically adjusting subsequent sampling intensity and direction, to guide regulatory authorities to sample intensity of different batches of risk types (the sampling intensity of high-risk batch is large, and the sampling intensity of low-risk batch is small).
[0017] In one of the embodiments, the present application provides an agricultural product quality risk early warning method based on multi-source data fusion, further comprising:
[0018] The risk probability and risk type output by risk assessment model are compared with the exact result data obtained by subsequent targeted laboratory sampling, and the comparison data (i.e. predicted value and true value) are used as new training sample, injected into risk assessment model training process, so as to continuously iterate and optimize the algorithm and parameters of risk assessment model, so that the prediction accuracy of risk assessment model is continuously improved over time.
[0019] In one of the embodiments, the present application provides an agricultural product quality risk early warning method based on multi-source data fusion, further comprising:
[0020] The risk assessment results (risk probability and risk type) are connected with the market mechanism. If the risk probability of all risk types of a batch of agricultural products is lower than a preset first threshold value (one risk type corresponds to one first threshold value), the batch of agricultural products is determined as high-quality grade agricultural products, an encrypted quality traceability code is automatically generated, and after the code is scanned by a consumer, the whole-chain data of the batch of agricultural products is displayed and compared with the national standard.
[0021] In one of the embodiments, the present application provides a multi-source data fusion based agricultural product quality risk early warning method, further comprising:
[0022] The whole-chain data of the agricultural products are disclosed to consumers, and different batches of agricultural products are set with different compensation prices based on the risk probability and risk type output by the risk assessment model. The compensation price represents the pre-committed compensation amount that the producer needs to pay to the consumer if the batch of products has quality safety problems (the higher the risk probability and the more serious the harm type, the compensation price increases exponentially). When the consumer decides to purchase, it is considered that the quality contract with clear compensation clauses is accepted.
[0023] In one of the embodiments, the present application provides a multi-source data fusion based agricultural product quality risk early warning method, further comprising:
[0024] The risk assessment results (risk probability and risk type) are connected with the market mechanism. If the risk probability of all risk types of a batch of agricultural products is lower than a preset first threshold value (one risk type corresponds to one first threshold value), the batch of agricultural products is determined as high-quality grade agricultural products, an encrypted quality traceability code is automatically generated, and after the code is scanned by a consumer, the whole-chain data of the batch of agricultural products is displayed and compared with the national standard.
[0025] In one of the embodiments, the present application provides a multi-source data fusion based agricultural product quality risk early warning system, comprising:
[0026] A multi-source data acquisition module is configured to construct a multi-source heterogeneous dataset. In a pre-production link, first data in the soil is monitored (which can be continuously monitored by Internet of Things sensors deployed in the field or periodically inspected by using low-cost, portable soil heavy metal rapid testers, pH test paper, etc.). The first data includes heavy metal (such as cadmium, lead, mercury, and arsenic) content and pH value. In a production link, second data is automatically recorded. The second data includes water quality reports of irrigation water sources, types and times of application of fertilizers and pesticides. In a post-production link, third data is collected. The third data includes harvesting time, temperature and humidity changes of storage warehouses, and environmental conditions during transportation. Meanwhile, microbial monitoring is performed on water sources and personnel hygiene that may be contacted in all links to obtain fourth data.
[0027] A multi-source data processing module is configured to clean, align, and correlate the multi-source heterogeneous data, convert the multi-source heterogeneous data into standardized data streams that can be used for analysis, match the first data, the second data, the third data, and the fourth data of the same batch of agricultural products through unified spatial identifiers (such as plot codes) and time stamps (such as batch numbers) as indexes, and construct a fusion database with the batch of agricultural products as the unique identifier.
[0028] A risk prediction module is configured to train a risk assessment model based on the fusion database by applying a machine learning algorithm, predict the risk probability of each batch of agricultural products facing various types of pollution (such as heavy metal enrichment, fungal toxin breeding, and microbial overlimit) through the risk assessment model, and use the risk probability and risk type output by the risk assessment model as the core basis for dynamically adjusting the subsequent inspection intensity and direction, and guide the regulatory department to inspect the inspection intensity of different batches of risk types (the inspection intensity of high-risk batches is large, and the inspection intensity of low-risk batches is small).
[0029] In one embodiment, the present application provides a multi-source data fusion-based agricultural product quality risk early warning system, which further comprises:
[0030] A model updating module is configured to compare the risk probability and risk type output by the risk assessment model with the exact result data obtained by subsequent targeted laboratory inspection, use the comparison data (i.e., predicted value and true value) as new training samples, and inject the new training samples into the risk assessment model training process, so as to continuously iterate and optimize the algorithm and parameters of the risk assessment model, and continuously improve the prediction accuracy of the risk assessment model over time.
[0031] In one embodiment, the present application provides a multi-source data fusion-based agricultural product quality risk early warning system, which further comprises:
[0032] The agricultural product data viewing module is used for connecting the risk assessment result (risk probability and risk type) with a market mechanism, and if the risk probability of all risk types of a batch of agricultural products is lower than a preset first threshold value (one risk type corresponds to one first threshold value), the batch of agricultural products is determined as high-quality grade agricultural products, and an encrypted quality traceability code is automatically generated, after a consumer scans the code, the full-chain data of the batch of agricultural products is displayed and compared with national standards.
[0033] In one of the embodiments, the application provides a multi-source data fusion based agricultural product quality risk early warning system, further comprising:
[0034] The quality contract construction module is used for disclosing the full-chain data of agricultural products to consumers, and setting different compensation prices for different batches of agricultural products based on the risk probability and risk type output by the risk assessment model, the compensation price represents a pre-agreed compensation amount (the higher the risk probability and the more serious the harm type, the compensation price exponentially increases) that the producer needs to pay to the consumer if the quality and safety problem occurs, and when the consumer decides to purchase, it is regarded as accepting the quality contract with clear compensation clauses.
[0035] In one of the embodiments, the application provides a multi-source data fusion based agricultural product quality risk early warning system, further comprising:
[0036] The automatic destruction module is used for connecting the risk assessment result (risk probability and risk type) with a market mechanism, if the risk probability of multiple risk types of a batch of agricultural products is higher than a preset second threshold value (one risk type corresponds to one second threshold value), and the economic value of the batch of agricultural products is lower than a preset value (such as feed corn), a predictive destruction program is started, the batch of agricultural products will not be harvested, and is directly plowed back into the soil as green manure, the whole process is monitored and verified through satellites and the Internet of Things, the behavior of actively avoiding pollution and resource waste is quantified, is identified as a carbon sequestration and methane emission reduction project (because the methane generated by grain mold and the carbon emissions in subsequent transportation and processing are avoided), and tradable carbon credits are generated.
[0037] Compared with the prior art, the application has the beneficial effects that: the application detects various pollutions such as pesticides, chemical pollutants other than pesticides, and biological pollution, and obtains risk types and risk probabilities through a risk assessment model, so as to adjust the detection intensity of different batches of risk types (avoid too many detection categories, time and effort), so that the detection result is more directional, and the root cause of high-risk agricultural products is detected in a shorter time. BRIEF DESCRIPTION OF DRAWINGS
[0038] Figure 1A first part flow diagram of a method for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0039] Figure 2 A second part flow diagram of a method for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0040] Figure 3 A third part flow diagram of a method for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0041] Figure 4 A fourth part flow diagram of a method for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0042] Figure 5 A fifth part flow diagram of a method for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0043] Figure 6 A first part diagram of a system for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0044] Figure 7 A second part diagram of a system for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0045] Figure 8 A third part diagram of a system for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0046] Figure 9 A fourth part diagram of a system for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application.
[0047] Figure 10 A fifth part diagram of a system for early warning of agricultural product quality risks based on multi-source data fusion is provided for the embodiment of the present application. DETAILED DESCRIPTION
[0048] In order to make the objectives, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.
[0049] It is to be understood that the terms "first", "second", and the like used herein can be used to describe various elements, but unless otherwise specifically stated, the elements thus described are not limited by these terms. These terms are only used to distinguish one element from another. For example, a first xx script can be termed a second xx script, and, similarly, a second xx script can be termed a first xx script, without departing from the scope of the present application.
[0050] In one embodiment, as shown in Figure 1 A multi-source data fusion-based agricultural product quality risk early warning method includes the following steps:
[0051] Step S1, a multi-source heterogeneous data set is constructed. In the pre-production link, the first data in the soil is monitored (the Internet of Things sensors deployed in the field can be used for continuous monitoring, or low-cost, portable soil heavy metal rapid testers, pH test paper, etc. can be used for regular inspection). The first data includes heavy metal (such as cadmium, lead, mercury, arsenic) content and pH value. In the production link, the second data is automatically recorded. The second data includes water quality reports of irrigation water sources, types and times of fertilizer and pesticide application. In the post-production link, the third data is collected. The third data includes harvesting time, temperature and humidity changes of storage warehouse, and environmental conditions during transportation. At the same time, the water sources and personnel hygiene that may be contacted in all links are monitored for microorganisms to obtain the fourth data.
[0052] Step S2, the multi-source heterogeneous data is cleaned, aligned and associated to convert into standardized data streams that can be used for analysis. Through unified spatial identification (such as plot code) and time stamp (such as batch number) as index, the first data, the second data, the third data and the fourth data of the same batch of agricultural products are matched to construct a fusion database with agricultural product batch as the unique identifier.
[0053] Step S3, based on the fusion database, a risk assessment model is trained by applying a machine learning algorithm. The risk assessment model is used to predict the risk probability of each batch of agricultural products facing various types of pollution (such as heavy metal enrichment, fungal toxin breeding, and microbial over-limit). The risk probability and risk type output by the risk assessment model are used as the core basis for dynamically adjusting the subsequent inspection intensity and direction, guiding the regulatory department to inspect the inspection intensity of different batches of risk types (the inspection intensity of high-risk batches is large, and the inspection intensity of low-risk batches is small).
[0054] In the prenatal link, the soil pollution census map disclosed by the environmental protection department and the water quality historical data of the water source from the water conservancy department can be integrated in advance to identify whether there is environmental pollution. Satellite or unmanned aerial vehicle remote sensing technology is used to macroscopically monitor the growth of crops and identify possible industrial pollution sources around. Then, Internet of Things sensors are deployed in the planting fields near the pollution sources to effectively reduce investment in the early stage and increase investment in the subsequent development (such as social development leading to a decrease in cost). Internet of Things sensors are also deployed in the environment-qualified areas to record the water quality of irrigation and report.
[0055] The risk assessment model is constructed and trained, multiple source data (such as heavy metal content, pH value, temperature and humidity time series change, and farming operation) are used as input features, and the pollution results confirmed by the laboratory (such as “heavy metal exceeding standard” and “mycotoxin positive”) are used as labeled targets. Then, a machine learning algorithm (such as gradient boosting decision tree GBDT or random forest) suitable for processing high-dimensional and heterogeneous data and capable of outputting probability is selected. The risk assessment model is trained through cross-validation to learn the complex nonlinear relationship between input features and output risks. Finally, the generated risk assessment model can calculate the probability of each pre-defined risk type to which the new batch data belongs, thereby realizing risk early warning.
[0056] In one embodiment, as shown in Figure 2 The method based on multi-source data fusion for agricultural product quality risk early warning further comprises:
[0057] Step S4, compare the risk probability and risk type output by the risk assessment model with the exact result data obtained by subsequent targeted laboratory sampling, and use the comparison data (i.e. predicted value and true value) as new training samples to inject the risk assessment model training process, thereby continuously iterating and optimizing the algorithm and parameters of the risk assessment model, so that the prediction accuracy of the risk assessment model is continuously improved over time.
[0058] For example, after the risk assessment model predicts that a batch of agricultural products has a 60% risk of aflatoxin, the regulatory department immediately conducts targeted laboratory sampling on the batch, and obtains the exact detection result (for example, the actual detection is positive). Then, the multi-source feature data of the batch (model input), the model prediction probability (0.6), and the laboratory true result (1, representing positive) are combined into a new training sample, which is injected into the training data set of the risk assessment model. In the next round of model iteration training, this new sample containing the prediction-actual comparison relationship will participate in the loss function calculation, and the gradient boosting decision tree GBDT will further adjust its tree structure and node split value by fitting the new sample, thereby correcting the model's judgment of the correlation between similar feature patterns (such as specific temperature and humidity changes, storage conditions) and the growth of aflatoxin, so that subsequent predictions are closer to the actual situation.
[0059] In one embodiment, as shown in Figure 3 The method for early warning of agricultural product quality risks based on multi-source data fusion further comprises:
[0060] Step S5: The risk assessment results (risk probability and risk type) are connected with the market mechanism. If the risk probability of all risk types of a batch of agricultural products is lower than a preset first threshold value (one risk type corresponds to one first threshold value), the batch of agricultural products is determined as high-quality agricultural products, and an encrypted quality traceability code is automatically generated. After the code is scanned by a consumer, the whole-chain data of the batch of agricultural products is displayed and compared with the national standards.
[0061] Meanwhile, high-risk producers can be included in the list of key supervision, and their risk records can be linked with agricultural credit and insurance preferential policies to encourage producers to improve the quality of agricultural products by using market means.
[0062] In one embodiment, as shown in Figure 4 The method for early warning of agricultural product quality risks based on multi-source data fusion further comprises:
[0063] Step S6: The whole-chain data of the agricultural products are disclosed to consumers, and different batches of agricultural products are set with different compensation prices based on the risk probability and risk type output by the risk assessment model. The compensation price represents the pre-committed compensation amount that the producer needs to pay to the consumer if the batch of products has quality and safety problems (the higher the risk probability and the more serious the harm type, the compensation price increases exponentially). When the consumer decides to purchase, it is considered that the consumer has accepted the quality contract with clear compensation clauses.
[0064] The setting of the compensation price can also be based on market supply and demand and consumer psychological expectations. If the compensation amount is too low, it cannot compensate for the potential health loss and trust cost of the consumer, and the consumer will refuse to buy due to insufficient protection, resulting in that the high-risk products are naturally eliminated by the market. On the contrary, if the compensation amount is too high, it can greatly attract consumers, but it will excessively squeeze the profit space of the producers, and even make them exit the market due to the unbearable insurance or guarantee cost. Therefore, the pre-committed compensation amount can be priced through market feedback, forcing the producers to make accurate cost-benefit accounting. The producers have to invest in cost to improve the quality to reduce the risk probability and pay a lower compensation guarantee, or pay a high premium due to low quality and face the risk of unsalable products, which encourages the producers to improve the quality of agricultural products.
[0065] In one embodiment, as shown in Figure 5 The method for early warning of agricultural product quality risks based on multi-source data fusion further comprises:
[0066] Step S7, the risk assessment results (risk probability and risk type) are connected with the market mechanism. If the risk probability of multiple risk types of a batch of agricultural products is higher than the preset second threshold value (one risk type corresponds to one second threshold value), and the economic value of the batch is lower than the preset value (such as feed corn), the predictive destruction program is started, the batch of crops will not be harvested, and will be directly plowed back into the soil as green manure. The whole process is monitored and verified by satellite and Internet of Things. This active pollution avoidance and resource waste behavior is quantified and identified as a carbon sequestration and methane emission reduction project (because the methane generated by grain mold and the subsequent carbon emissions during transportation and processing are avoided), and tradable carbon credits are generated.
[0067] Carbon credits can be sold to enterprises with emission reduction needs (such as airlines and energy companies) in domestic and international carbon trading markets, thereby obtaining additional cash income to partially compensate for economic losses caused by not harvesting crops; producers can also choose to retain these credits to offset carbon emissions generated by their own farm operations or other production processes, helping them achieve carbon neutrality goals, enhance their environmental image, and prepare for future carbon tax policies.
[0068] In one embodiment, as shown in Figure 6 a multi-source data fusion-based agricultural product quality risk early warning system includes:
[0069] A multi-source data acquisition module 1 is used to construct a multi-source heterogeneous data set. In the pre-production link, the first data in the soil is monitored (low-cost, portable soil heavy metal rapid testers, pH test paper, etc. can be used for regular inspections, or Internet of Things sensors deployed in the field can be used for continuous monitoring). The first data includes heavy metal (such as cadmium, lead, mercury, and arsenic) content and pH value. In the production link, the second data is automatically recorded, including water quality reports of irrigation water sources, types and times of fertilizer and pesticide application. In the post-production link, the third data is collected, including harvesting time, temperature and humidity changes in storage warehouses, and environmental conditions during transportation. At the same time, microorganism monitoring is performed on all links that may come into contact with water sources and personnel hygiene to obtain the fourth data.
[0070] A multi-source data processing module 2 is used to clean, align, and associate multi-source heterogeneous data, and convert them into standardized data streams that can be used for analysis. Through unified spatial identifiers (such as plot codes) and timestamps (such as batch numbers) as indexes, the first data, second data, third data, and fourth data of the same batch of agricultural products throughout the whole chain are matched to construct a fusion database with agricultural product batches as the unique identifier.
[0071] The risk prediction module 3 is used for training a risk assessment model based on the fusion database by applying a machine learning algorithm, predicting the risk probability of each batch of agricultural products facing various types of pollution (such as heavy metal enrichment, fungal toxin breeding, and microbial overlimit) through the risk assessment model, and taking the risk probability and risk type output by the risk assessment model as the core basis for dynamically adjusting the subsequent sampling intensity and direction, thereby guiding the regulatory department to adjust the sampling intensity of different batches of risk types (the sampling intensity of high-risk batches is large, and the sampling intensity of low-risk batches is small).
[0072] The multi-source data acquisition module 1 constructs a risk perception foundation with a global perspective by collecting multi-dimensional data covering the whole chain of agricultural products before, during and after production. This design breaks the limitations of traditional single-link detection and realizes the risk traceability capability from the source to the table.
[0073] Due to the heterogeneity of multi-source data (such as sensor data, manual records, and report files), direct analysis will lead to information fragmentation. The multi-source data processing module 2 dynamically correlates the whole chain data through spatial identification (plot code) and time stamp (batch number) to form a standardized data stream in batches. This design not only solves the problem of agricultural data fragmentation, but also builds a unique digital identity throughout production, logistics and market, making it possible to track the risk of each batch of agricultural products and laying a data foundation for subsequent accurate modeling.
[0074] The traditional sampling has blindness and hysteresis, and the risk prediction module 3 can dynamically guide the regulatory department to optimize the sampling strategy through the risk probability and type output by the risk assessment model, so as to focus on the supervision of high-risk batches and reduce the intervention of low-risk batches, thereby improving the allocation efficiency of regulatory resources.
[0075] In one embodiment, as shown in FIG. 1, Figure 7 The agricultural product quality risk early warning system based on multi-source data fusion further comprises:
[0076] The model updating module 4 is used for comparing the risk probability and risk type output by the risk assessment model with the exact result data obtained by the subsequent targeted laboratory sampling, taking the comparison data (i.e. predicted value and true value) as new training samples, and injecting the new training samples into the risk assessment model training process, so as to continuously iterate and optimize the algorithm and parameters of the risk assessment model, and continuously improve the prediction accuracy of the risk assessment model over time.
[0077] The initial prediction of the risk assessment model may be biased. By introducing laboratory sampling real data and comparing the prediction results, a closed-loop learning mechanism of prediction, verification and feedback is formed. This design enables the risk assessment model to continuously absorb new knowledge and continuously correct the understanding of the risk correlation law in the complex agricultural system, so as to adapt to dynamic conditions such as environmental changes and planting mode adjustments, and ensure that the risk early warning ability is more and more accurate and reliable over time.
[0078] In one embodiment, as shown in Figure 8 Agricultural product quality risk early warning system based on multi-source data fusion also includes:
[0079] The agricultural product data viewing module 5 is used for connecting the risk assessment results (risk probability and risk type) with the market mechanism. If the risk probability of all risk types of a batch of agricultural products is lower than a preset first threshold value (one risk type corresponds to one first threshold value), the batch of agricultural products is determined as high-quality agricultural products, and an encrypted quality traceability code is automatically generated. After the consumer scans the code, the full-chain data of the batch of agricultural products is displayed and compared with the national standard.
[0080] The invisible quality data is converted into visual trust credentials through the traceability code. Consumers are allowed to scan the code to view the full-chain data of agricultural products and compare them with the national standards. This design not only builds consumer confidence through transparency, but also reduces the brand building cost of high-quality agricultural products through the automatic mechanism of passing the standard as certification, and promotes the transformation of the market to high-quality and high-price positive incentives.
[0081] In one embodiment, as shown in Figure 9 Agricultural product quality risk early warning system based on multi-source data fusion also includes:
[0082] The quality contract construction module 6 is used for disclosing the full-chain data of agricultural products to consumers, and setting different compensation prices for different batches of agricultural products based on the risk probability and risk type output by the risk assessment model. The compensation price represents the pre-committed compensation amount that the producer needs to pay to the consumer if the batch of products has quality and safety problems (the higher the risk probability and the more serious the harm type, the compensation price increases exponentially). When the consumer decides to buy, it is considered to accept the quality contract with clear compensation clauses.
[0083] This will unarguably anchor the quality and safety responsibility on the production end. The compensation risk will form a strong economic pressure mechanism, forcing producers to improve quality control levels from the source throughout the process in order to avoid potential huge compensation, thereby transforming traditional government supervision pressure into direct market economic pressure.
[0084] In one embodiment, as shown in Figure 10As shown, the agricultural product quality risk early warning system based on multi-source data fusion further comprises:
[0085] The automatic destruction module 7 is used for connecting the risk assessment result (risk probability and risk type) with the market mechanism. If the risk probability of a batch of agricultural products of multiple risk types is higher than the preset second threshold value (one risk type corresponds to one second threshold value), and the economic value of the batch is lower than the preset value (such as feed corn), a predictive destruction program is started, the batch of crops will not be harvested, and is directly plowed back into the soil as green manure. The whole process is monitored and verified through satellites and the Internet of Things. This active pollution avoidance and resource waste behavior is quantified, identified as a carbon sequestration and methane emission reduction project (because the methane generated by grain mold and the subsequent carbon emissions during transportation and processing are avoided), and tradable carbon credits are generated.
[0086] For high-risk and low-economic-value agricultural products (such as feed corn), a predictive destruction carbon credit acquisition mechanism is innovatively designed. The unharvested crops are directly plowed back into the soil through satellite and Internet of Things monitoring, thereby avoiding mold methane and reducing transportation and processing carbon emissions, and generating tradable carbon credits. This design converts negative value risk products into positive value environmental assets, which not only prevents pollution from flowing into the market, but also provides partial compensation for producers through the carbon trading market, forming a sustainable risk disposal path that achieves economic and environmental benefits.
[0087] It should be understood that although each step in the flowchart of each embodiment of the present application is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least a part of the steps in each embodiment can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.
[0088] Each technical feature of the above-described embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-described embodiments are not described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present disclosure.
[0089] The above embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but should not be understood as a limitation on the patent scope of the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.
[0090] The above only describes the preferred embodiments of the present application, and is not used to limit the present application, and any modifications, equivalent replacements and improvements, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.
[0091] In addition, it should be understood that, although the present specification is described in terms of embodiments, not every embodiment contains only one independent technical solution, and the description manner of the specification is only for clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be combined to form other embodiments which can be understood by those skilled in the art.
Claims
1. A method for early warning of agricultural product quality risk based on multi-source data fusion, characterized in that, The agricultural product quality risk early warning method based on multi-source data fusion comprises the following steps: Construct a multi-source heterogeneous data set, monitor the first data in the soil in the pre-production link, the first data including heavy metal content, pH value; in the production link, automatically record the second data, the second data including water quality report of irrigation water source, application type and time of chemical fertilizer and pesticide; in the post-production link, collect the third data, the third data including harvesting time, temperature and humidity change of storage warehouse, environmental conditions in the transportation process; at the same time, the water source and personnel hygiene that may be contacted in all links are monitored for microorganisms to obtain the fourth data; The multi-source heterogeneous data is cleaned, aligned and associated, and is converted into standardized data flow that can be used for analysis, the first data, the second data, the third data and the fourth data of the same batch of agricultural products are matched through unified spatial identification and time stamp as index, and a fusion database taking the batch of agricultural products as the unique identification is constructed; Based on the fusion database, a risk assessment model is trained by applying a machine learning algorithm, the risk probability of each batch of agricultural products facing various types of pollution is predicted through the risk assessment model, the risk probability and risk type output by the risk assessment model are used as the core basis for dynamically adjusting the subsequent sampling intensity and direction, and the supervision department is guided to sample the different batches of risk types. 2.The agricultural product quality risk early warning method based on multi-source data fusion according to claim 1, characterized in that, Further comprising: The risk probability and risk type output by the risk assessment model are compared with the exact result data obtained by the subsequent targeted laboratory sampling, and the comparison data are used as new training samples and injected into the risk assessment model training process, so that the algorithm and parameters of the risk assessment model are continuously iterated and optimized, and the prediction accuracy of the risk assessment model is continuously improved over time. 3.The agricultural product quality risk early warning method based on multi-source data fusion according to claim 1, characterized in that, Further comprising: The risk assessment result is connected with the market mechanism, if the risk probability of all risk types of a batch of agricultural products is lower than a preset first threshold, the batch of agricultural products is determined as high-quality grade agricultural products, an encrypted quality traceability code is automatically generated, and after the consumer scans the code, the full-chain data of the batch of agricultural products is displayed and compared with the national standard.
4. The agricultural product quality risk early warning method based on multi-source data fusion according to claim 1, characterized in that, Further comprising: The full-chain data of the agricultural products is disclosed to the consumers, and different batches of agricultural products are set different compensation prices based on the risk probability and risk type output by the risk assessment model, the compensation price represents the pre-committed compensation amount that the producer needs to pay to the consumer if the quality safety problem of the batch of products occurs, and when the consumer decides to purchase, it is regarded as accepting the quality contract with clear compensation clauses.
5. The agricultural product quality risk early warning method based on multi-source data fusion according to claim 3 or 4, characterized in that, Further comprising: The risk assessment result is connected with the market mechanism, if the risk probability of all risk types of a batch of agricultural products is lower than a preset first threshold, the batch of agricultural products is determined as high-quality grade agricultural products, an encrypted quality traceability code is automatically generated, and after the consumer scans the code, the full-chain data of the batch of agricultural products is displayed and compared with the national standard.
6. An agricultural product quality risk early warning system based on multi-source data fusion, characterized in that, Further comprising: A multi-source data acquisition module is configured to construct a multi-source heterogeneous data set, monitor first data in the soil in a pre-production link, and monitor second data in a production link. The first data includes heavy metal content and pH value. The second data includes water quality reports of irrigation water sources, types and times of application of fertilizers and pesticides. Third data is collected in a post-production link. The third data includes harvesting time, temperature and humidity changes of storage warehouses, and environmental conditions in the transportation process. Microbial monitoring of water sources and personnel hygiene that may be contacted in all links is performed to obtain fourth data. A multi-source data processing module is configured to clean, align and associate the multi-source heterogeneous data, convert the multi-source heterogeneous data into standardized data streams that can be used for analysis, match the first data, the second data, the third data and the fourth data of the same batch of agricultural products through unified spatial identifiers and time stamps as indexes, and construct a fusion database with the batch of agricultural products as the unique identifier. A risk prediction module is configured to train a risk assessment model based on the fusion database and apply a machine learning algorithm. The risk assessment model is used to predict the risk probability of each batch of agricultural products facing various types of pollution. The risk probability and risk type output by the risk assessment model are used as the core basis for dynamically adjusting the subsequent inspection intensity and direction, and are used to guide the supervision department to inspect the inspection intensity of different batches of risk types.
7. The agricultural product quality risk early warning system based on multi-source data fusion according to claim 6, characterized in that, Further comprising: A model updating module is configured to compare the risk probability and risk type output by the risk assessment model with the exact result data obtained by subsequent targeted laboratory inspection, and use the comparison data as new training samples to inject the risk assessment model training process, thereby continuously iterating and optimizing the algorithm and parameters of the risk assessment model, and continuously improving the prediction accuracy of the risk assessment model over time. 8.The agricultural product quality risk early warning system based on multi-source data fusion according to claim 6, characterized in that, Further comprising: An agricultural product data viewing module is configured to connect the risk assessment result with the market mechanism. If the risk probability of all risk types of a batch of agricultural products is lower than a preset first threshold, the batch of agricultural products is determined to be a high-quality grade agricultural product, and an encrypted quality traceability code is automatically generated. After the consumer scans the code, the full-chain data of the batch of agricultural products is displayed and compared with the national standard. 9.The agricultural product quality risk early warning system based on multi-source data fusion according to claim 6, characterized in that, Further comprising: A quality contract construction module is configured to disclose the full-chain data of the agricultural products to consumers, and set different compensation prices for different batches of agricultural products based on the risk probability and risk type output by the risk assessment model. The compensation price represents the pre-committed compensation amount that the producer needs to pay to the consumer if the batch of products has quality and safety problems. When the consumer decides to purchase, it is considered that the consumer has accepted the quality contract with clear compensation clauses.
10. The agricultural product quality risk early warning system based on multi-source data fusion according to claim 8 or 9, characterized in that, Further comprising: An automatic destruction module is used to connect the risk assessment results with the market mechanism. If the risk probability of multiple risk types of a batch of agricultural products is higher than the preset second threshold value, and the economic value of the batch is lower than the preset value, the predictive destruction program is started. The batch of crops will not be harvested and will be directly plowed back into the soil as green manure. The whole process is monitored and verified by satellite and Internet of Things. This active pollution avoidance and resource waste avoidance behavior is quantified as a carbon sequestration and methane emission reduction project to generate tradable carbon credits.