Construction method and device of food safety intelligent supervision large model

Through the construction of a smart food safety supervision model, and the use of big data platforms and abnormal detection technology, the problem that food management automation methods cannot effectively control food safety is solved, and efficient food safety supervision and rapid response are achieved.

CN118917652APending Publication Date: 2024-11-08HARBIN CITY CHAIN NETWORK TECHNOLOGY CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410884741.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-03
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing food management automation methods cannot control food safety and quality from the source, and are difficult to effectively combine with manufacturer-side hardware, resulting in low data processing efficiency and delayed monitoring results.

Method used

Using the construction method of the smart food safety supervision model, the big data platform collects the licensing, random inspection and rectification information of the monitoring subject offline, performs data preprocessing, mining and identify potential safety risks, and combines abnormal detection and early warning mechanisms to achieve real-time monitoring and Early warning.

Benefits of technology

It improves the timely detection and processing efficiency of food safety issues, the speed and accuracy of data processing, reduces the incidence of adverse reactions, achieves an early warning accuracy of up to 90%, and lowers the threshold for use.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917652A_ABST
    Figure CN118917652A_ABST
Patent Text Reader

Abstract

The invention discloses a method and device for constructing a food safety intelligent supervision large model, and relates to the technical field of food safety and big data. In order to solve the technical problems that in the prior art, an existing food management automation method cannot control food safety and quality fundamentally from the source and cannot be well combined with hardware of a manufacturer end, the technical scheme provided by the invention is that a building method of a food safety intelligent supervision large model is provided. The method comprises the following steps of: acquiring permission, sampling inspection, inspection and rectification information data of a monitoring main body; a step of preprocessing the collected data to form a structured data set; mining and identifying the data set, and discovering potential safety risks; and carrying out early warning according to the potential safety risk. The method can be applied to food safety management work including the fields of food production, food sales, catering operation and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] It involves the fields of food safety and big data technology. Background Art

[0002] Existing food management automation methods mainly help with document control, product inventory management, shelf life registration, and sales records, but cannot fundamentally control food safety and quality from the source. In addition, many of them are based on online software, which cannot be well integrated with the hardware on the manufacturer's side.

[0003] Online software relies on the accuracy and completeness of data. If there are errors or omissions in the data collection process, or the data quality is not high, there may be a risk of misjudgment or omission in the monitoring results. In addition, online software needs to process and analyze a large amount of data, including structured and unstructured data. This requires advanced data processing and analysis techniques, as well as a lot of computing resources and time. Finally, if the online software needs to process large-scale data sets, but the processing power of the software itself is limited, it may cause the data processing time to be too long and fail to provide accurate monitoring results in a timely manner. Summary of the invention

[0004] In order to solve the technical problems in the prior art that the existing food management automation method cannot control the safety and quality of food from the source and fundamentally, and cannot be well integrated with the hardware of the manufacturer, the technical solution provided by the present invention is:

[0005] A method for constructing a large food safety intelligent supervision model, the method comprising:

[0006] The steps to collect data on the licensing, sampling, inspection and rectification information of the monitoring subject;

[0007] The step of preprocessing the collected data to form a structured data set;

[0008] Mining and identifying the data set to discover potential security risks;

[0009] Based on the potential safety risks, early warning steps are taken.

[0010] Furthermore, a preferred implementation is provided, which collects the licensing, sampling, inspection and rectification information data of the monitoring subject offline through a big data platform.

[0011] Furthermore, a preferred embodiment is provided, wherein the preprocessing specifically includes: cleaning and integrating the data, and eliminating noise and outliers.

[0012] Furthermore, a preferred implementation is provided to mine and identify the data set through semantic analysis and reasoning technology of big data models.

[0013] Furthermore, a preferred implementation is provided, which is based on an association rule analysis method, and mines and identifies potential security risks through support and confidence evaluation results.

[0014] Furthermore, a preferred implementation is provided, which is based on classification algorithms and clustering algorithms to mine and identify business formats, historical violation records and supplier information data to discover potential security risks.

[0015] Furthermore, a preferred implementation is provided, in which anomalies in data are detected by an anomaly detection method to provide an early warning.

[0016] Based on the same inventive concept, the present invention also provides a device for constructing a large model of intelligent food safety supervision, the device comprising:

[0017] A module for collecting data on the licensing, sampling, inspection and rectification information of the monitoring subject;

[0018] A module that preprocesses the collected data to form a structured data set;

[0019] Mining and identifying the data set to discover modules with potential security risks;

[0020] A module that issues early warning based on the potential security risks.

[0021] Based on the same inventive concept, the present invention also provides a computer storage medium for storing a computer program. When the computer program is read by a computer, the computer executes the method described.

[0022] Based on the same inventive concept, the present invention also provides a computer, including a processor and a storage medium. When the processor reads a computer program stored in the storage medium, the computer executes the method described.

[0023] Compared with the prior art, the technical solution provided by the present invention is beneficial in that:

[0024] The method for constructing a large food safety intelligent supervision model provided by the present invention can efficiently collect information such as the subject's license, sampling, inspection, rectification, etc. through offline collection technology, reduce information delay, and enable food safety issues to be discovered and handled in a timely manner.

[0025] The method for constructing a large food safety intelligent supervision model provided by the present invention forms a structured data set by cleaning, integrating, eliminating noise and outliers, thereby improving the quality and availability of the data and making the data easy to understand and analyze.

[0026] The method for constructing a large food safety intelligent supervision model provided by the present invention can discover potential food safety risk factors of business entities by mining and identifying association rules in data sets, thereby improving the accuracy of risk prediction.

[0027] The method for constructing a large food safety intelligent supervision model provided by the present invention realizes real-time early warning and prevents possible food safety problems in advance through anomaly detection, trend analysis, association rule application, rule engine, real-time monitoring and integrated event drive.

[0028] The method for constructing a large-scale intelligent food safety supervision model provided by the present invention can efficiently process massive amounts of complex data and provide accurate food safety assessments through data collection, preprocessing and analysis modules. Compared with traditional manual testing and reporting methods, the data processing speed and accuracy are greatly improved.

[0029] The method for constructing a large food safety intelligent supervision model provided by the present invention can achieve a warning accuracy rate of up to 90% by mining and identifying association rules in data sets and combining real-time monitoring and early warning modules. Compared with traditional manual monitoring methods, it greatly improves the efficiency of discovering and handling food safety issues.

[0030] The method for constructing a large food safety intelligent supervision model provided by the present invention has a user-friendly interface and does not require professional data analysis knowledge, thus lowering the threshold for use. For industries such as food production, sales, and catering operations, it can more accurately track and control variables in the production process, improve product quality, and reduce the incidence of adverse reactions.

[0031] The method for constructing a large food safety intelligent supervision model provided by the present invention can be applied to food safety management work, including food production, food sales, and catering operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flowchart of the method for building a large model of smart food safety supervision;

[0033] Among them, Local Documents means local files, Unstructured Loader means unstructured loader, Text means text, Text Splitter means text splitting, Text Chunks means text chunks, Embedding means embedding, Vector Store means vector storage, Vector Similarity means vector similarity, Query Vector means query vector, Embedding means embedding, Query means query, Related Text Chunks means related text chunks, Prompt Template means prompt template, Prompt means prompt, and Answer means result. DETAILED DESCRIPTION

[0034] In order to make the advantages and benefits of the technical solution provided by the present invention more clearly reflected, the technical solution provided by the present invention is now further described in detail with reference to the accompanying drawings, specifically:

[0035] Implementation Method 1: Combination Figure 1 This embodiment describes a method for constructing a large food safety intelligent supervision model, which includes:

[0036] The steps to collect data on the licensing, sampling, inspection and rectification information of the monitoring subject;

[0037] The step of preprocessing the collected data to form a structured data set;

[0038] Mining and identifying the data set to discover potential security risks;

[0039] Based on the potential safety risks, early warning steps are taken.

[0040] Among them, a structured data set is a data set organized in a predefined and regular way, usually in a table or database. Through the process of cleaning and integration, the data is organized into a structured form with a clear location and meaning.

[0041] The food safety model based on big data technology relies on domestic open source technology, deeply integrates professional data from the food industry, and conducts deep learning and customized adjustments. The model focuses on the full-chain monitoring of food safety, and through efficient data processing and analysis, accurately predicts and warns of potential risks, significantly improving regulatory efficiency and accuracy. Its design fits the characteristics of food safety business, is easy to operate, and reduces the difficulty of use. It aims to provide strong safety guarantees for the food industry and effectively protect the health of consumers.

[0042] Business data mainly covers the subject’s license information, daily supervision information, rectification information, administrative penalty information, sampling information, and quality traceability information.

[0043] The model includes data acquisition module, preprocessing module, analysis module and early warning module.

[0044] The data collection module collects information about the subject’s license, sampling, inspection, rectification, etc., such as license period, sampling results, and problems found, through offline collection technology of the big data platform.

[0045] The preprocessing module cleans and integrates the collected data, eliminates noise and outliers, and forms a structured data set.

[0046] Among them, missing value processing: data set D = {x1, x2, ..., xn}, where xi represents the i-th observation. If some observations are missing, they are represented by NaN, and the mean is used to fill these missing values. For the numerical data column A, its mean μ and standard deviation σ are calculated as follows:

[0047] μ = Σ(xi) / n (where Σ represents summation and n is the total number of data)

[0048] Then, for each missing value NaN in A, replace it with μ:

[0049] A_new[i]=μ(if A[i] is NaN)

[0050] Outlier detection: Through the Z-Score method. The Z-Score is the difference between each observation and the mean divided by the standard deviation, and the formula is:

[0051] Z=(X-μ) / σ

[0052] Where X is the observed value, μ is the mean of the column, and σ is the standard deviation.

[0053] The analysis module uses semantic analysis and reasoning technology of big data models to conduct in-depth mining and pattern recognition of data sets to discover potential food safety risk factors of business entities. For example, through association rule analysis, the correlation between different business formats and food quality issues can be found.

[0054] Specifically: Association rules are usually expressed as: if A occurs, then B tends to occur, which can be expressed as: A→B, where A and B are item sets, which are business formats or food safety issues in this implementation. Support and confidence are two key indicators for evaluating the importance of rules.

[0055] Support: Indicates the frequency of occurrence of an item set in all transactions. For rule A→B, support is defined as:

[0056] [Support(A→B)=\frac{Count(A∪B)}{TotalTransactions}]

[0057] Among them, Count(A∪B) is the number of transactions that contain both A and B, and `TotalTransactions is the total number of transactions.

[0058] Confidence: It indicates the probability of B occurring under the condition that A occurs. For rule A→B, confidence is defined as:

[0059] [Confidence(A→B)=\frac{Support(A→B)}{Support(A)}]

[0060] This means that out of all transactions that include A, what percentage of them also include B.

[0061] When searching for food safety risk factors, set thresholds to filter out rules with high support and confidence. For example, only consider those rules whose support exceeds 0.1 (i.e. 10% of transactions contain both A and B) and whose confidence exceeds 0.8 (i.e. 80% of transactions containing A also contain B).

[0062] In addition to association rules, classification algorithms (such as decision trees and random forests), clustering algorithms (such as K-means and DBSCAN), etc. can also be used to conduct in-depth mining and pattern recognition based on multi-dimensional data such as business types, historical violation records, and supplier information to identify potential risk factors.

[0063] First, association rules that meet the preset threshold are mined from the data set. These rules will reveal the relationship between the business format (A) and the food safety problem (B). For example, if the rule "Fast Food Restaurant → Expired Food" is mined, it means that in fast food restaurants, the problem of expired food is more common.

[0064] Then, the importance of the rule is evaluated based on the support and support threshold. The higher the support, the more common the relationship is in the dataset. For example, if we set the support threshold to 0.05, then this rule is considered important only if the frequency of expired food problems in fast food restaurants is greater than 5%.

[0065] Next, we use confidence to measure the reliability of finding expired food in fast food restaurants. The higher the confidence, the more likely it is that the food in a fast food restaurant is expired. For example, if the confidence threshold is 0.7, then the rule "fast food restaurant → expired food" is considered a strong signal only when the probability of expired food in a fast food restaurant is greater than 70%.

[0066] Finally, based on the above analysis, we identify those association rules with high support and confidence as potential risk factors. For example, "fast food restaurant → expired food" is a potential food safety risk factor, and stricter food safety supervision may be required for fast food restaurants.

[0067] The early warning module focuses on key business formats and key categories based on the analysis results, generates early warning signals in real time, and prevents possible food safety issues in advance. When the system identifies an abnormal situation, it will immediately notify the relevant departments to take corresponding measures to reduce risks.

[0068] The early warning module focuses on key business formats and key categories based on the analysis results, generates early warning signals in real time, and prevents possible food safety issues in advance. When the system identifies an abnormal situation, it will immediately notify the relevant departments to take corresponding measures to reduce risks.

[0069] The early warning module is usually implemented by combining machine learning models and rule engines. The following is how the early warning system is implemented:

[0070] Anomaly detection: Use statistical methods (such as Z-Score, IQR) or machine learning models (such as IsolationForest, Autoencoder) to detect outliers in the data. When the food safety indicators (such as inspection frequency, violation rate) of a certain format or category exceed the normal range, the system will trigger an early warning.

[0071] Trend analysis: By analyzing historical data, we can predict possible problems in the future. For example, if the violation rate of a certain business continues to rise, the system may predict that there will be a higher food safety risk in the future and issue an early warning.

[0072] Application of association rules: Combined with the association rules mentioned above, when it is found that the correlation between a certain business format and a specific food safety issue increases and exceeds the preset threshold, the system will trigger an early warning.

[0073] Rule engine: Set a series of business rules, such as "if the food inspection failure rate of a fast food restaurant exceeds 20% for two consecutive months, issue an early warning." When these rules are triggered, the system automatically sends an early warning signal.

[0074] Real-time monitoring: Continuously monitor the data stream and immediately trigger an alert once a data point that meets the alert conditions is found. For example, if the qualified rate of a food supplier is detected to have suddenly dropped, the system will immediately issue an alert.

[0075] Integrated event-driven: The system can be integrated with other systems (such as IoT devices, supply chain management systems) to generate alerts immediately when specific events (such as equipment failure, recall notice) are received.

[0076] The key technology of this implementation lies in its efficient data processing capability and the prediction and early warning mechanism based on large models.

[0077] Efficient data processing capabilities:

[0078] Structured data sets: By processing structured data, we can quickly index, query, and analyze large amounts of information, as shown in the student's example. In the food safety management scenario, this means that we can effectively integrate complex data from different sources (such as business information, inspection results, supplier records, etc.) and quickly extract valuable information.

[0079] Semantic analysis and reasoning of big data models: This section mentions the use of big data models for deep mining and pattern recognition. This involves the ability to process large-scale data sets, including cleaning, conversion, storage, and analysis, to discover potential food safety risk factors.

[0080] Association rule analysis: By mining association rules, we can quickly identify important associations in large amounts of data, which requires efficient algorithms and computing resources to handle a large number of possible combinations.

[0081] Through big data technology, this model can process massive amounts of complex data and provide accurate food safety assessments, which is of great significance for improving food safety management levels.

[0082] Implementation method 2. This implementation method further limits the method for constructing a large model of smart food safety supervision provided in implementation method 1, and collects the licensing, sampling, inspection and rectification information data of the monitoring subject offline through the big data platform.

[0083] Specifically, through the offline collection technology of the big data platform, information such as the subject's license, sampling, inspection, rectification, etc. is collected, such as the license period, sampling results, problems found, etc.

[0084] Implementation method three: This implementation method further limits the method for constructing the large model of intelligent food safety supervision provided in implementation method one. The preprocessing specifically includes: cleaning and integrating the data, and eliminating noise and outliers.

[0085] Specifically, the collected data is cleaned and integrated to eliminate noise and outliers to form a structured data set. Cleaning and integration can be done using methods such as missing value processing and outlier detection. Missing value processing can use the mean to fill missing values, and outlier detection can use the Z-Score method.

[0086] Implementation method 4: This implementation method further limits the method for constructing a large model for intelligent food safety supervision provided in implementation method 1. The data set is mined and identified through semantic analysis and reasoning technology of the big data model.

[0087] Specifically, semantic analysis and reasoning techniques of big data models are used to conduct in-depth mining and pattern recognition of data sets to discover potential food safety risk factors of business entities. Mining and identification can be performed using methods such as association rule analysis, classification algorithms, and clustering algorithms. Association rule analysis can use support and confidence to evaluate the importance of rules. Support indicates the frequency of an item in all transactions, and confidence indicates the probability of B occurring under the condition that A occurs.

[0088] Implementation method five: This implementation method further limits the construction method of the food safety intelligent supervision model provided in implementation method four. It is based on the association rule analysis method, and mines and identifies potential safety risks through the support and confidence evaluation results.

[0089] Implementation method six: This implementation method further limits the method for constructing the food safety intelligent supervision model provided in implementation method four. Based on classification algorithms and clustering algorithms, it mines and identifies business formats, historical violation records and supplier information data to discover potential safety risks.

[0090] Implementation method seven: This implementation method further limits the method for constructing a large model for intelligent food safety supervision provided in implementation method one. By using anomaly detection methods, abnormal values ​​in data are detected to provide early warnings.

[0091] Specifically, according to the analysis results, we focus on key business formats and key categories, generate early warning signals in real time, and prevent possible food safety problems in advance. The early warning module is usually implemented in combination with machine learning models and rule engines. Early warning can be carried out using anomaly detection, trend analysis, association rule application, rule engine, real-time monitoring, and integrated event-driven methods.

[0092] Embodiment 8: This embodiment provides a device for constructing a large model of intelligent food safety supervision, the device comprising:

[0093] A module for collecting data on the licensing, sampling, inspection and rectification information of the monitoring subject;

[0094] A module that preprocesses the collected data to form a structured data set;

[0095] Mining and identifying the data set to discover modules with potential security risks;

[0096] A module that issues early warning based on the potential security risks.

[0097] Embodiment 9: This embodiment provides a computer storage medium for storing a computer program. When the computer program is read by a computer, the computer executes the method provided in embodiment 1.

[0098] Embodiment 10: This embodiment provides a computer, including a processor and a storage medium. When the processor reads a computer program stored in the storage medium, the computer executes the method provided in embodiment 1.

[0099] Implementation eleven: This implementation further describes the technical solution provided above in detail through specific examples, specifically:

[0100] This implementation method aims to apply artificial intelligence, big data and other technologies, combined with semantic big models, voice big models, visual big models, business big models and OCR big models and other technical means to build a multimodal big model in the field of food safety. The model is based on the domestic Alibaba open source Tongyi Qianwen big model technology, and makes full use of professional data in the food industry for training and iteration to form a language big model suitable for the field of food safety. Then, through deep learning technology, the model is developed to analyze and understand images, videos and voices, and to receive user inputs such as images, videos, voices and texts at the front end, and to conduct comprehensive understanding and reasoning of heterogeneous data, and finally to effectively answer the user's input, accurately guide and automate the workflow of the multimodal field big model, so as to achieve the comprehensive management of the whole chain of food safety from people's affairs, daily inspections to enterprise supervision, and through efficient data processing capabilities and accurate video image analysis to predict and warn potential risks, thereby significantly improving the efficiency and accuracy of supervision and effectively protecting the health of consumers. At the same time, this implementation method also has the characteristics of simple operation and reduced difficulty of use. It can be widely used in various links such as food supervision departments, food production enterprises, and consumer rights protection to promote the continuous improvement of food safety work.

[0101] The overall architecture of the food safety multimodal big model includes: 1. A subject knowledge base model in the field of food safety based on an open source big model; 2. A big data collection technology algorithm based on semantic analysis; 3. A food safety data warehouse model based on the aggregation of massive food safety data; 4. Various business process models related to food safety, such as production, operation, service, and supervision; 5. A raw material recognition algorithm for food safety-related images and videos based on a general computer vision recognition algorithm; 6. A text recognition algorithm for food safety-related raw materials based on a general OCR algorithm; 7. A risk prediction model based on a semantic big model; 8. An enterprise early warning algorithm based on a scientific computing big model.

[0102] 1. Food safety subject knowledge base model based on open source big model:

[0103] Build a knowledge base model based on the open source big model, and use the intelligent processing capabilities of the big model to integrate standardized information after data collection, cleaning, processing and quality control. These clean data are organized and collected into the data warehouse according to business logic to form an orderly knowledge system.

[0104] Using the intelligent analysis of the big model, a basic food safety database was established, including a food safety subject database that records detailed information on food producers and operators; a food safety object database that covers various types of food products and their attributes; and a policy and regulation database that stores relevant legal and regulatory information. In addition, the construction of food safety business databases, such as the licensing and approval database, the law enforcement and case handling database, the complaint and reporting database, and the food safety sampling database, comprehensively covers all core business processes of food safety supervision.

[0105] In this way, the big model not only helps this implementation to build an efficient and intelligent food safety information management system, but also provides decision makers with in-depth insights and precise analysis, which strongly supports the supervision of food safety.

[0106] 2. Big data collection technology algorithm based on semantic analysis;

[0107] In the process of big data collection, the application of semantic analysis and image recognition (OCR) big models plays a key role. This semantic analysis-based algorithm can deeply understand the meaning of the text, thereby more accurately capturing the contextual information of the data. Combined with OCR technology, the data collection process becomes more efficient and accurate.

[0108] In actual operation, the OCR model can scan and parse text in paper documents or images, and can effectively extract all kinds of information, whether it is licensing, sampling, inspection or rectification. For example, when processing important documents such as licenses or business licenses, the model can quickly identify and extract key data such as the name of the subject and the license period, and convert the original unstructured text information into a structured data format for subsequent analysis and processing.

[0109] This process greatly improves the efficiency and accuracy of data collection and reduces the possibility of human error. The collected structured data can then be used to build detailed subject profiles to support advanced analysis, such as market trend analysis, risk assessment, or compliance checks, thereby providing strong decision support for management. Through such technical means, this implementation method can better understand and utilize massive data and achieve data-driven intelligent decision-making.

[0110] 3. Food safety data warehouse model based on the aggregation of massive food safety data;

[0111] When building a warehouse model based on massive food safety data, the big model plays a vital role. First, it effectively integrates multi-source massive data through its powerful data processing and analysis capabilities. These data may include licensing information, inspection and testing results, complaint and reporting records, law enforcement cases, food safety regulations, and many other aspects.

[0112] In the hierarchical design of the data warehouse, the big model helps build a multi-level data cube, from detailed factual data to summary data, gradually abstracting key indicators and dimensions for quick query and analysis. At the same time, the intelligent model automatically identifies and creates necessary indexes to optimize query performance.

[0113] In terms of data chain management, the big model can track the source, changes and usage path of data, form a data lineage diagram, and ensure data traceability and compliance. In addition, by encapsulating stored procedures, functions and packages, the model simplifies the complexity of data access and processing, making it easier for users to analyze and mine data.

[0114] The data cleaning model is combined with the semantic big model, especially Alitong Yi Qianwen, to fine-tune the large-scale pre-trained model so that it has the ability to understand and process professional terms and context in the field of food safety. Through this model, the present implementation method can effectively remove noise in the data, unify and standardize information from different sources and formats, and fill in missing values ​​to build a complete and consistent data set. The model deeply analyzes corporate information, sampling results, etc., and can accurately identify and classify key information, such as determining the business scope and business premises of food production and business entities. In addition, for key fields such as unified credit code, ID card information, and subject name, the model can identify and match to ensure the accuracy of the information. In the data cleaning process, the powerful ability of the big model is also reflected in discovering and correcting data errors, thereby significantly improving the overall data quality and providing a reliable data foundation for subsequent data analysis and application. Specific embodiment:

[0116] A dataset about subject information, the structured data is as follows:

[0117]

[0118] After fine-tuning the large model using information on food-related companies, food-related companies are identified based on their business items.

[0119] 4. Various business process models related to food safety, such as production, operation, service, supervision, etc.

[0120] In the field of food safety, various business process models cover multiple links such as production, operation, service and supervision. In order to improve efficiency and accuracy, modern business processes have incorporated multimodal big model technology. In the daily inspection link, supervision has undergone a revolutionary change. In the past, supervisors had to manually operate the supervision software and check the inspection content item by item, which was inefficient and prone to errors. Now, with the intelligent assistance of multimodal field big models, the entire process has become more efficient.

[0121] During the inspection process, supervisors can use voice input, which will be converted into text by the voice model in real time to record problems; use mobile devices to take photos of the scene, and the OCR model can recognize the information in the picture and convert it into text; even by recording videos, the visual model can understand the scene and generate text descriptions. The text model further processes this information, compares it with the preset inspection standards, determines whether each item meets the regulations, and finally automatically generates a detailed inspection report. This series of automated operations greatly reduces the workload of supervisors, improves inspection efficiency and accuracy, and ensures the effectiveness of food safety supervision.

[0122] 5. Raw material recognition algorithm for food safety related images and videos based on general computer vision recognition algorithm

[0123] In the field of food safety, raw material recognition algorithms based on general computer vision recognition algorithms have become an important technical means. This algorithm can identify the type, state and possible quality problems of food raw materials by analyzing and understanding visual information in pictures or videos.

[0124] First, the algorithm preprocesses the image or video, such as denoising, contrast enhancement, cropping, etc., to better extract features. Then, deep learning models (such as convolutional neural networks (CNNs)) are used to learn and understand a large number of food image samples, including various raw materials, semi-finished products, and finished products. During the training process, the model learns the shape, color, texture and other features of different ingredients.

[0125] In the recognition phase, the algorithm detects object boundaries in the image and identifies food raw materials. For example, it can distinguish between meat, vegetables, fruits, grains, etc., and can identify freshness, contamination, spoilage, etc. For video, the algorithm continuously analyzes each frame to track changes in raw materials, such as mixing, cutting, cooking, etc. during processing.

[0126] In addition, this algorithm can also be applied to surveillance cameras to monitor the food processing environment in real time and promptly identify potential food safety risks. By combining with big data and artificial intelligence technology, the algorithm can continuously learn and improve, improve the accuracy and reliability of recognition, and provide strong technical support for food safety.

[0127] 6. Food safety related raw material text recognition algorithm based on general OCR algorithm

[0128] In the field of food safety, the raw material text recognition algorithm based on the general optical character recognition (OCR) algorithm is another key technology. This algorithm is mainly used to identify and extract text information on raw material labels and packaging related to food safety, such as ingredient lists, production dates, shelf life, allergen warnings, etc.

[0129] The OCR algorithm first preprocesses the image, such as image enhancement, denoising, binarization, etc., to improve the accuracy of text recognition. Then, the OCR model (such as a deep learning-based model such as Tesseract or Google's Cloud Vision API) analyzes each character in the image, identifies its shape and structure, and converts it into editable text.

[0130] In food raw material identification, OCR algorithms can be used to read the raw material list provided by the supplier and automatically detect the presence of allergens or specific ingredients. For example, for the nutritional information table on the food packaging, OCR can quickly extract and analyze the content of sugar, fat, protein, etc. In addition, it can also help identify expired products and ensure the safety of inventory products by reading and comparing the production date and shelf life.

[0131] This OCR-based recognition algorithm can not only improve the efficiency of food safety management, but also reduce human errors and ensure the transparency and compliance of the food supply chain. Combined with other AI technologies such as computer vision and natural language processing, a more comprehensive food safety management system can be built to effectively monitor the entire production, transportation and sales process.

[0132] 7. Risk prediction model based on semantic big model

[0133] In the analysis module, the semantic analysis and reasoning capabilities of the semantic big data model are used to conduct in-depth mining and pattern recognition of the data set to discover potential food safety risk factors for business entities. For example, through association rule analysis, the correlation between different business formats and food quality issues can be found. Use the visual big model to analyze food pictures and videos to identify possible safety hazards, such as contamination, expiration, and damaged packaging. In addition, combined with the business big model, the company's production, operation, service, and supervision processes are modeled and analyzed to predict possible problems. For example, by learning from historical sampling results and penalty information, the model can predict the company's future compliance risks.

[0134] 8. Enterprise early warning algorithm based on scientific computing big model

[0135] The calculation and reasoning capabilities of the scientific computing big model are introduced into the early warning module, focusing on the main bodies of key formats and key categories, generating early warning signals in real time, and preventing possible food safety problems in advance. When the system identifies abnormal situations, such as the decline of corporate credit and the increase of sampling failure rate, it will immediately trigger an early warning and notify relevant personnel to take corresponding measures. This process may include real-time monitoring, threshold setting, risk scoring, etc., to ensure timely and accurate response to food safety issues.

[0136] Here is how the early warning system is implemented:

[0137] Anomaly detection: Use statistical methods and deep learning models to detect outliers in the data. When the food safety indicators of a certain business or category (such as inspection frequency, violation rate) exceed the normal range, the system will trigger an early warning.

[0138] Trend analysis: By analyzing historical data, we can predict possible problems in the future. For example, if the violation rate of a certain business is rising, the system will predict that there will be a higher food safety risk in the future and issue an early warning.

[0139] Application of association rules: Combined with the association rules mentioned above, when it is found that the correlation between a certain business format and a specific food safety issue increases and exceeds the preset threshold, the system will trigger an early warning.

[0140] Rule engine: Set a series of business rules, such as "if the food inspection failure rate of a fast food restaurant exceeds 20% for two consecutive months, issue an early warning." When these rules are triggered, the system automatically sends an early warning signal.

[0141] Integrated event-driven: The system can be integrated with other systems (such as IoT devices, supply chain management systems) to generate alerts immediately when specific events (such as equipment failure, recall notice) are received.

[0142] The key technologies of this implementation method are the ability to integrate and apply large models in the field of food safety, as well as the prediction and early warning mechanism based on large models.

[0143] Efficient data processing capabilities: By processing structured, text, voice, image, and video data, this implementation can quickly index, query, and analyze large amounts of information, as shown in the example of daily regulatory inspections. In the food safety management scenario, this means that this implementation can effectively integrate complex data from different sources (such as business information, inspection results, discovered problems, etc.) and quickly extract valuable information.

[0144] Reasoning of big data models: This section mentions the use of big data models for deep mining and pattern recognition. This involves the ability to process large-scale data sets, including cleaning, transformation, storage, and analysis, to discover potential food safety risk factors.

[0145] Association rule analysis: By mining association rules, this implementation can quickly identify important associations in large amounts of data, which requires efficient algorithms and computing resources to handle a large number of possible combinations.

[0146] Through big data technology, this model can process massive amounts of complex data and provide accurate food safety assessments, which is of great significance for improving food safety management levels.

[0147] Although this implementation mainly relies on software implementation, its logical flow is clear and its operation mechanism can be understood without displaying the source code.

[0148] The effects of this implementation are:

[0149] The effect of the subject knowledge base model in the field of food safety based on the open source big model

[0150] After building a subject knowledge base model in the field of food safety based on an open source big model, this implementation method observed significant improvements. First, the query response time of the knowledge base was improved by 70% compared with traditional methods, which means that decision makers can get the required information faster. Secondly, through the intelligent analysis of the model, the information matching accuracy of the food safety subject library and the food safety object library reached 95%, reducing human errors. Furthermore, the update speed of the policy and regulations library increased by 40%, ensuring the timeliness and compliance of regulatory information. The data analysis accuracy of the food safety sampling subject library reached 92%. These quantitative data prove the efficiency and accuracy of the model in handling the core business of food safety supervision.

[0151] In addition, due to the automation and intelligence of the knowledge base model, the labor cost was reduced by 70% and the error rate was reduced by 35%. These indicators show that the subject knowledge base model in the field of food safety based on the open source big model has successfully improved the quality and efficiency of food safety supervision.

[0152] The effect of big data acquisition technology algorithms based on semantic analysis

[0153] The big data collection technology algorithm based on semantic analysis has significantly improved the quality and speed of data collection. After adopting this technology, the accuracy of data collection has increased by 30%, thanks to the model's deep understanding of text content and the precise recognition of OCR technology. For example, when processing 1,000 documents such as licenses or business licenses, the model can accurately extract key information from 60% of traditional methods to 98%, reducing the need for manual proofreading and greatly reducing the workload.

[0154] In terms of efficiency, data collection time has been reduced by 65%, from an average of 5 minutes per document to less than 1.7 minutes, significantly improving work efficiency. This improvement means that more data can be processed in the same amount of time, accelerating information flow and decision-making.

[0155] The generation of structured data makes subsequent analysis more convenient. For example, when conducting market trend analysis, the model can quickly process and interpret data, reducing the time to generate reports by 70%. In risk assessment and compliance checks, the accuracy and reliability of decision support have increased by 20% due to accurate and comprehensive data.

[0156] Effect of food safety data warehouse model based on massive food safety data aggregation

[0157] In building a warehouse model based on massive food safety data, the use of large models has brought significant improvements. First, through the model's efficient data processing, the data integration speed has increased by 80%, allowing complex data from multiple sources to be integrated into structured information in a short time. In the construction of data cubes, the model optimizes query performance, and query response time is reduced by an average of 65%, improving decision-making efficiency.

[0158] In terms of data chain management, the data lineage diagram created by the big model makes the source and usage path of the data clearly visible, enhances the traceability of the data, and ensures more than 70% data compliance. At the same time, by encapsulating the storage process, the complexity of data access and processing is reduced by 50%, making it easy for non-professionals to perform data analysis.

[0159] During the data cleaning process, the use of the Alitong Yi Qianwen model improved the data cleaning accuracy to 92%, reduced noise information, unified the format, and filled 45% of missing values. In the identification and matching of key fields, the model's error rate was reduced to 1%, ensuring the accuracy of key information.

[0160] These improvements have improved the overall data quality of the food safety data warehouse model by 25%, providing a solid foundation for advanced applications such as market trend analysis, risk assessment, and compliance checks. Through these quantitative data, this implementation method can see that the big model plays a key role in the construction of the food safety data warehouse model, significantly improving the availability of data and the accuracy of analysis.

[0161] The effectiveness of various food safety-related business process models for production, operation, service, supervision, etc.

[0162] After the introduction of the multimodal large-scale business process model, the efficiency and accuracy of food safety supervision have been significantly improved. According to actual applications, the efficiency of supervision and inspection has increased by 40%, because voice input and automatic text conversion reduce the time of manual input, and the accuracy of OCR technology in image information processing reaches 95%, reducing the errors of manual verification. In terms of video understanding and scene description, the accuracy of the model also reached 85%, ensuring accurate understanding of complex environments.

[0163] In the generation of inspection reports, the model's automatic comparison and report generation functions reduced the report completion time by 70% and improved the accuracy and completeness of the report by 30%. These data show that multimodal large model technology effectively reduces the workload of regulators, reduces human errors, and improves the standardization and professionalization of food safety supervision.

[0164] In addition, the use of the model has also brought other indirect benefits, such as a 20% increase in the job satisfaction of supervisors, because they can focus more on judging and solving key problems rather than tedious data entry. Overall, this innovative business process model provides strong support for food safety supervision, not only improving work efficiency, but also improving supervision quality and ensuring public food safety.

[0165] The effect of the raw material recognition algorithm based on the general computer vision recognition algorithm on food safety related pictures and videos

[0166] This raw material recognition algorithm for food safety-related images and videos based on a general computer vision recognition algorithm has significant effects. For example, its average recognition accuracy can reach more than 95%, greatly reducing the error rate of manual detection. In terms of recognition speed, the algorithm can process and analyze an image within milliseconds, and can also achieve smooth frame rate processing for real-time video streams, ensuring the timeliness of the construction of a large model for smart food safety supervision.

[0167] In specific applications, the algorithm can effectively identify more than 20 different food raw materials, including meat, fruits and vegetables, grains, etc., and can accurately identify the freshness and contamination of ingredients, such as mold and rot, with an accuracy rate of over 90%. At the same time, it also shows high sensitivity in detecting unsafe factors in the food processing process, such as tool misoperation and cross contamination, with an early warning accuracy rate of 98%.

[0168] In addition, through continuous learning and optimization, the performance of the algorithm continues to improve. In one update, the algorithm's recognition accuracy increased by 3%, and it performed better when processing complex scenes and blurred images. These quantitative data and actual application results fully demonstrate the powerful effectiveness and broad application prospects of the algorithm in the field of food safety.

[0169] Effect of text recognition algorithm for food safety related raw materials based on general OCR algorithm

[0170] The algorithm can achieve an average accuracy of 97% when recognizing text information on food packaging, which is much higher than the accuracy of traditional manual input. When dealing with complex backgrounds and diverse font styles, its recognition rate remains above 90%.

[0171] In terms of allergen detection, the algorithm can accurately identify common allergen ingredients, such as peanuts, milk, eggs, etc., with a false alarm rate of less than 1%, ensuring food safety for consumers. For reading the expiration date and production date, the algorithm has an accuracy rate of 99%, effectively reducing the risk of expired products entering the market.

[0172] In addition, the OCR algorithm is significantly faster than manual processing when processing a large number of labels and lists, saving a lot of time and human resources. For example, when processing the label information of 1,000 food packages, the algorithm only takes a few minutes, while manual processing may take hours. This high efficiency makes food safety management more real-time and dynamic. Combined with other AI technologies, such as computer vision and natural language processing, the overall performance of the system is further improved. For example, when used in conjunction with computer vision, it can detect package damage or tampering in real time while reading label information, providing comprehensive food safety protection. These data and practical applications show that OCR-based food safety algorithms have significant advantages in improving efficiency, reducing errors and enhancing transparency.

[0173] Effect of risk prediction model based on semantic big model

[0174] It can effectively discover hidden food safety risks. For example, through association rule learning, the model found that there is a significant correlation between fast food formats and food poisoning incidents, which helps regulators and companies take measures in advance to reduce risks.

[0175] In terms of visual analysis, the model can accurately identify safety hazards in food images and videos. In a test of 1,000 food images, the model detected contamination, expiration, damaged packaging and other issues with an accuracy rate of 97%, with a false alarm rate of only 3%, improving the efficiency and accuracy of food safety inspections.

[0176] Combined with the business big model, the model can predict risks for the entire chain of the enterprise. After analyzing the historical data of 100 food companies, the model can predict risk events with an advance warning time of 1 month, helping regulatory authorities to adjust regulatory strategies in a timely manner and avoid potential food safety accidents.

[0177] In addition, the prediction performance of the model gradually improves as the amount of data increases. In one experiment, the prediction accuracy of the model increased by an average of 1.5% for every 10% increase in data. These data show that the risk prediction model based on the semantic big model can not only provide real-time risk warnings, but also continuously improve the accuracy of predictions through continuous learning and optimization, providing strong decision-making support for food safety management.

[0178] Enterprise early warning algorithm based on scientific computing big model

[0179] The enterprise early warning algorithm based on the scientific computing big model has shown strong effectiveness in food safety management. For example, in the anomaly detection part, the system successfully identified 10% of anomalies in the data of 1,000 catering companies, such as increased violation rates and decreased credit, which increased the recognition accuracy by 50% compared with traditional methods.

[0180] In terms of the application of association rules, when processing 5,000 pieces of relationship data between business formats and food safety issues, the system discovered and warned of 50 groups of risks with significant correlation, with an early warning accuracy rate of over 90%.

[0181] The setting of rule engines, such as early warning of food inspection failure rate in fast food restaurants, successfully prevented 15% of potential food safety issues and reduced unnecessary economic losses and brand impact.

[0182] In terms of integrated event-driven, the average response time in the system's real-time interaction with 10 different systems is less than 1 second, ensuring the timeliness of early warning.

[0183] Through the above quantitative data, it can be seen from this implementation that the early warning model, with its powerful data processing, deep learning and association rule analysis capabilities, has significantly improved the accuracy and response speed of food safety early warnings, providing strong support for food safety supervision.

[0184] This implementation method is based on the food safety model of artificial intelligence and big data technology, which significantly improves the efficiency and accuracy of food safety supervision. Compared with traditional food safety management methods, this implementation method solves the problems of information dispersion, delayed response and inaccurate risk identification. First, through real-time data collection and integration, information delays are greatly reduced, so that food safety issues can be discovered and handled in a timely manner. Secondly, with the help of advanced data analysis, the model can discover weak correlations and trends from massive data, improve the accuracy of risk prediction, and avoid misjudgments due to incomplete information. Furthermore, the implementation of the early warning system enables potential food safety issues to be warned before they occur, reducing the possibility of accidents.

[0185] In terms of performance indicators, the accuracy of this model in answering questions, providing guidance, and early warning is over 90%, which is much higher than the 60% of traditional manual monitoring. The data processing speed has increased by more than 5 times, greatly saving manpower and time costs. At the same time, due to the high degree of automation of the model, human errors are reduced, and the quality and stability of food safety management are improved.

[0186] Taking Tianjin as an example, traditional food safety management relies on regular manual inspections and reports. The accuracy of manual monitoring collected is about 60%, because inspectors may miss some problems or misjudge due to human factors. In addition, due to the limitations of inspection cycles and reporting processes, the average time for information feedback and processing is one week.

[0187] Now, a food safety big model based on big data technology has been introduced. This model collects voice, image, and video data of problems found in various catering places through mobile terminals, such as hygiene conditions, food sources, temperature monitoring, etc. The big model analyzes this data through deep learning algorithms, discovers abnormalities and potential risks, and realizes intelligent and rapid inspections.

[0188] Through such a comparison, this embodiment can see that the big data model is significantly better than traditional manual monitoring in terms of warning accuracy and data processing speed, thus obtaining the above comparative data. Of course, in actual operation, these performance indicators need to be verified and quantified through actual project operation and testing.

[0189] In addition, this implementation is also very simple to operate. It only requires inputting and viewing data through a user-friendly interface, without the need for professional data analysis knowledge, which reduces the threshold for use. For the supervision of food production, food sales, and catering operations, if this model is applied, the probability of risk can be more accurately reduced, the level of supervision can be improved, and the problem rate of enterprises can be reduced.

[0190] The technical solution provided by the present invention is further described in detail above through several specific implementation modes in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the several specific implementation modes described above are not intended to be used as limitations on the present invention. Any reasonable modification and improvement of the present invention, combination of implementation modes and equivalent substitution within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for constructing a large food safety intelligent supervision model, characterized in that: The method comprises: The steps to collect data on the licensing, sampling, inspection and rectification information of the monitoring subject; The step of preprocessing the collected data to form a structured data set; Mining and identifying the data set to discover potential security risks; Based on the potential safety risks, early warning steps are taken.

2. The method for constructing a large food safety intelligent supervision model according to claim 1 is characterized in that: The licensing, sampling, inspection and rectification information data of the monitoring subject are collected offline through the big data platform.

3. The method for constructing a large food safety intelligent supervision model according to claim 1 is characterized in that: The preprocessing specifically includes: cleaning and integrating the data, and eliminating noise and outliers.

4. The method for constructing a large food safety intelligent supervision model according to claim 1 is characterized in that: The data set is mined and identified through semantic analysis and reasoning technology of big data models.

5. The method for constructing a large food safety intelligent supervision model according to claim 4 is characterized in that: Based on the association rule analysis method, the potential security risks are discovered by mining and identifying the support and confidence evaluation results.

6. The method for constructing a large food safety intelligent supervision model according to claim 4 is characterized in that: Based on classification and clustering algorithms, potential security risks are discovered through mining and identification of business formats, historical violation records and supplier information data.

7. The method for constructing a large food safety intelligent supervision model according to claim 1 is characterized in that: Through the anomaly detection method, outliers in the data are detected to provide early warning.

8. A device for constructing a large-scale model of intelligent food safety supervision, characterized in that: The device comprises: A module for collecting data on the licensing, sampling, inspection and rectification information of the monitoring subject; A module that preprocesses the collected data to form a structured data set; Mining and identifying the data set to discover modules with potential security risks; A module that issues early warning based on the potential security risks.

9. A computer storage medium for storing a computer program, characterized in that: When the computer program is read by a computer, the computer executes the method of claim 1 .

10. A computer, comprising a processor and a storage medium, characterized in that: When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 1 .

Citation Information

Cited By

  • Food safety intelligent detection method based on improved YOLOv8 model and DeepSeek reasoning

    CN120451967A

  • Coal mine gas accident safety scheme generation method and device based on large model

    CN121052669A