A method and system for multi-source data analysis, tracing, and early warning of plant oils

By combining unique oil sample coding with hyperspectral detection, the problem of unstructured data analysis and fusion in the vegetable oil industry chain has been solved, realizing automated, real-time, and precise traceability of oil information and improving the speed of food safety emergency response.

CN123070980APending Publication Date: 2026-11-10HUNAN ACAD OF FORESTRY +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611531578.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-30
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In existing technologies, unstructured data in the vegetable oil industry chain is difficult to automatically analyze and integrate, and there is a lack of unified correlation between oil information across different links. Traditional testing cycles are long, making it impossible to achieve real-time on-site traceability and second-level early warning, and failing to meet the needs of rapid response to food safety incidents.

Method used

The method combines unique oil sample coding with hyperspectral detection. Oil sample codes are generated using QR code or barcode technology. Mobile scanning technology is used to automatically convert the collected information into structured data. Hyperspectral oil rapid detection equipment is used to achieve real-time on-site detection. Data analysis is performed using AI multimodal models to generate intelligent evaluation reports.

Benefits of technology

It has achieved automated, real-time, and precise traceability of oil information, overcome the bottleneck of low efficiency in unstructured data parsing, broken down cross-source information silos, realized on-site real-time monitoring and second-level early warning, and improved the speed of emergency response to food safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070980A_ABST
    Figure CN123070980A_ABST
Patent Text Reader

Abstract

This invention relates to a method and system for multi-source data analysis, traceability, and early warning of vegetable oils. It generates a unique code for each oil sample through an intelligent system, uses on-site scanning to bind the location and time information of the collection point, and employs a hyperspectral oil rapid detection device to generate the spectral characteristics of the oil sample in real time. After the collected oil sample is delivered to the laboratory, it is scanned into the database and the test results are imported, establishing an oil source information database and a spectral feature database. Consumers can trace the origin, processing, composition, and batch information of the vegetable oil by scanning the code, while regulatory authorities can verify the vegetable oil in real time using hyperspectral equipment. This invention uses an AI multimodal model combined with XGBoost and SHAP algorithms to achieve quantitative risk scoring and intelligent report generation for vegetable oils, and uses a mathematical model to inversely deduce the mixing ratio of each component in blended oils. This invention achieves real-time on-site traceability and early warning for the entire vegetable oil supply chain, providing consumers and regulatory authorities with an efficient and accurate technical tool.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of food safety testing and quality monitoring technology, and in particular to a method and system for multi-source data analysis, traceability, and early warning of vegetable oils. Background Technology

[0002] The quality and safety of vegetable oils are directly related to public health. Quality is typically assessed based on three main dimensions: oil type, production process, and origin. Traditional vegetable oil regulatory models rely primarily on information collection, laboratory testing, and expert analysis, but these methods face the following technical bottlenecks: The efficiency of unstructured data parsing and recombination is low. In the vegetable oil industry chain, a large amount of heterogeneous data such as enterprise paper ledgers, customs declarations, and inspection reports is difficult to automatically convert into analyzable structured information, resulting in low data entry efficiency. Cross-source information correlation is difficult; information in the industrial chain is scattered across multiple links such as planting, import, production, circulation, and supervision, lacking a unified correlation system. The correlation between oil products and their indicators, attributes, and test results in the circulation process lacks on-site verification mechanisms, resulting in serious information silos. The emergency response is delayed; traditional testing cycles are long, usually taking several hours to several days, and are mostly post-event sampling inspections, lacking on-site, real-time (second-level) traceability, monitoring and early warning capabilities, which cannot meet the needs of rapid response to food safety incidents.

[0003] In existing technologies, hyperspectral detection technology has been applied to the field of oil product testing, but it is mostly limited to laboratory environments and has not been effectively integrated with the whole industry chain data system and artificial intelligence algorithms. At the same time, existing traceability systems are mostly based on single QR code or barcode technology, which is difficult to cope with the accurate traceability needs in complex scenarios such as oil sample cross-selling, mixing, and transportation contamination. Summary of the Invention

[0004] In view of this, the present invention provides a method and system for multi-source data analysis, traceability and early warning of vegetable oils, which solves the technical problems in the existing vegetable oil industry chain, such as the difficulty in automatically parsing and integrating unstructured data, the lack of unified correlation and on-site verification mechanism for oil information across links, and the inability to achieve real-time on-site traceability monitoring and second-level early warning due to the long testing cycle of traditional laboratories. It provides a technical solution for the whole-chain supervision of vegetable oils by combining unstructured data acquisition, transformation, multi-source data fusion and AI real-time analysis.

[0005] To achieve the above objectives, this invention provides a method for source tracing and early warning of multi-source data analysis of plant oils, comprising the following steps: S1. Oil Sample Code Generation: Generate a unique code UID for the oil sample based on the oil sample type, customer type, collection type, and collection batch quantity. Output the code as a QR code or barcode and affix it to the oil sample container. The oil sample code is bound to the database. S2. By scanning the identification code on the oil sample collection container with an authorized mobile terminal, the location information of the collection point, time information, collector's identity, collection purpose and oil sample collection container are bound and uploaded to the cloud platform to establish an oil source information database; S3. At the collection site, use a hyperspectral oil rapid detection device to perform spectral scanning on the collected oil sample to obtain spectral feature data. After associating the spectral feature data with the oil sample code, store it in the spectral feature database. S4. When the oil sample arrives at the laboratory and is stored, the identification code on the oil sample container is scanned and collected by a smart mobile terminal to register the oil sample for storage. S5. Oil Sample Traceability: Based on the oil sample code and spectral characteristic data, record and trace the entire process of oil sample from collection, storage, testing, production process to delivery. S6. Input the data from the oil source information database and the spectral feature database into the AI ​​multimodal model for intelligent analysis, predict, assess, warn, and evaluate the quality, safety, mixing degree, and safety risks of the oil sample, and generate an intelligent assessment report.

[0006] Preferably, the authorized mobile terminal is equipped with an authorized APP or WeChat mini-program. The collector completes the binding by scanning the identification code, and the collector supplements the source information of the collected oil sample as needed; the location information of the collection point includes latitude and longitude coordinates.

[0007] Preferably, laboratory testing personnel scan the QR code on the container of the oil sample to be tested using a smart mobile terminal to bind the oil sample to the testing equipment, testing time, and testing steps, and directly import the equipment test results into the database; The process of oil samples arriving at the laboratory for storage also includes: scanning or inputting paper test reports; when the tested oil sample is selected as the modeling oil sample for the hyperspectral rapid oil detection equipment, the laboratory test results are input into the modeling system of the hyperspectral rapid oil detection equipment as the dependent variable of the hyperspectral feature model, updating the spectral quantitative analysis model, and realizing the quantitative analysis of this type of oil sample collected on-site by the hyperspectral rapid oil detection equipment through sample accumulation.

[0008] Preferably, the process of oil sample delivery to the laboratory for storage also includes oil sample splitting and testing, and binding the sample splitting code at the time of oil sample splitting and testing with the oil sample code upon entering the storage.

[0009] Preferably, oil sample traceability includes two modes: consumer traceability and regulatory verification; The consumer traceability specifically includes: consumers scanning the traceability code affixed to the oil container with a mobile terminal to obtain information including place of origin, key indicators of production process, batch number, production time, and laboratory test report; The regulatory verification specifically includes: regulatory personnel extracting oil samples from oil containers, using high-spectral oil rapid detection equipment to conduct on-site real-time detection to obtain spectral characteristics, quantitatively monitoring the oil sample composition, tracing back to the batch number at the factory, and identifying the type of oil.

[0010] Preferably, the AI ​​multimodal model intelligent analysis includes blended oil ratio estimation: the oil database collects the fatty acid composition and quality indicators of each raw oil, and uses a mathematical model to deduce the estimated ratio of each component in the blended oil.

[0011] Preferably, the AI ​​multimodal model intelligent analysis of oil safety risk prediction, assessment, early warning and health assessment specifically includes: inputting oil traceability information, oil sample component list image and consumer input information as AI multimodal data into the big language model, and extracting structured feature data, including acid value, peroxide value, brand, place of origin and batch; The structured feature data is input into the XGBoost gradient boosting decision tree model to obtain the risk score and risk level; The SHAP algorithm was used to calculate the marginal contribution of each feature in the risk score and to perform attribution analysis. The risk score, risk level, and SHAP attribution results are input into the large language model to generate an assessment report that includes natural language explanations, quantitative data, and traceability evidence.

[0012] This invention also provides a multi-source data analysis, tracing, and early warning system for plant oils, comprising: The oil sample coding generation module is used to generate a unique code for an oil sample and output it in the form of a QR code or barcode. The mobile terminal data acquisition module is used to scan the QR code and bind the data acquisition point information, time, collector, and acquisition purpose with the oil sample code. The hyperspectral detection module is used to acquire spectral characteristic data of oil samples in real time on site. The cloud platform database module is used to store the oil source information database and the spectral feature database, and to establish the relationship between the two; The laboratory testing data import module is used to import test results, equipment information, and paper reports into the database. The traceability module enables full-process tracing based on oil sample codes and spectral characteristics. The AI ​​multimodal analysis module is used to estimate the proportion of mixed oils and to predict, assess, and provide early warnings about the quality, safety, mixing degree, and safety risks of oil samples, generating intelligent assessment reports.

[0013] Compared with the prior art, the beneficial effects of the present invention are: This invention combines unique oil sample coding with mobile scanning technology to automatically transform heterogeneous information such as paper ledgers and handwritten labels that rely on manual processing into structured digital information. The data is collected and automatically entered, and the latitude and longitude, time, collection purpose and collector are automatically locked, which fundamentally breaks through the bottleneck of low efficiency in the parsing and reconstruction of unstructured data. This invention establishes a dual indexing mechanism for oil samples consisting of QR code encoding and spectral features. The spectral features serve as a permanent fingerprint for the oil sample, allowing it to be traced even if the oil sample itself deteriorates and becomes ineffective. This enables information such as origin, process, composition, and test reports that are scattered across various stages of planting, importing, production, factory exit, circulation, and supervision to be automatically aggregated through a unified index, completely breaking down cross-source information silos and solving the long-standing problem of lack of on-site correlation verification of information in the industrial chain. This invention enables real-time on-site detection with results within five to six seconds using a portable hyperspectral rapid detection device, transforming the traditional testing delay mode of several hours to several days into an on-site instant monitoring mode. Once an indicator is abnormal, an early warning can be triggered, greatly improving the speed of food safety emergency response. This invention constructs an intelligent analysis system that integrates information input, algorithm models, and expert systems. Through mathematical models, it can deduce the mixing ratio of various raw material oils in blended oil while obtaining routine test data, solving the pain point that laboratories cannot determine the degree of mixing. Furthermore, it utilizes XGBoost and SHAP algorithms to achieve risk quantification scoring and attribution analysis, providing professional decision support that surpasses human experience. This invention also fully leverages the fusion and guiding role of AI multimodal large models, unifying regulations, standards, scientific research literature, and oil traceability information to output intelligent assessment reports containing natural language descriptions, quantitative data, and traceability evidence. Simultaneously, it serves consumers and regulatory authorities through two methods: QR code traceability and hyperspectral on-site verification, respectively, achieving full-chain traceability from farmland or import to the table. This provides a comprehensive technical solution for the quality and safety of vegetable oils that combines real-time performance, accuracy, and operability. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention; Figure 2 A flowchart illustrating the process of establishing an oil source information database for this invention; Figure 3 A flowchart for establishing a spectral feature database for this invention; Figure 4 This is a flowchart illustrating the oil traceability process of this invention. Figure 5This is a schematic diagram illustrating the insertion of an oil sample into a cuvette and the insertion of the device into the hyperspectral oil rapid detection equipment according to an embodiment of the present invention; Figure 6 This is a flowchart of the multivariate data analysis and early warning model and health assessment algorithm for edible oil safety in an embodiment of the present invention. Detailed Implementation

[0015] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0016] Example 1: Addressing the challenges of safe circulation and regulation of vegetable oils, this embodiment addresses the issues from the digital collection of oil samples and the acquisition of spectral characteristics on-site. It then integrates on-site sample information with spectral data as a source of information for developing model algorithms and providing a tool for real-time quality assessment. This tool not only obtains monitoring results in real time but also integrates on-site photography of monitored oil sample indicators, automatically collects and merges regulations, standards, laboratory results, application cases, and relevant literature, and completes real-time analysis and early warning. The entire process of vegetable oil assessment is output in the form of intelligent reports. Furthermore, it can simulate risk quantification analysis provided by industry experts, simultaneously determining whether oil samples have been mixed and calculating the possible proportion of oil mixtures while testing oil sample components in the laboratory.

[0017] like Figure 1 As shown in this embodiment, a method for multi-source data analysis, tracing, and early warning of vegetable oils is provided. The aim is to establish a method for acquiring, transforming, and automatically collecting unstructured data and multi-source data using unique digital codes and spectral characteristics of oil samples. This allows the intelligent system to connect with AI algorithm models to analyze, predict, and interpret the acquired data. The intelligent system implements its system logic through a cloud platform and a hyperspectral oil rapid detection device. The intelligent system cloud platform can also be considered as application cloud platform software, including a database. Specifically, the method includes the following steps: S1. Oil Sample Code Generation: Generate a unique code UID for the oil sample based on the oil sample type, customer type, collection type, and collection batch quantity. Output the code as a QR code or barcode and affix it to the oil sample container. The oil sample code is bound to the database. In this embodiment, the intelligent system provides a software-based input interface for generating oil sample codes, allowing users to input the oil sample type, customer type, collection type, and collection batch quantity. The system generates a unique code (UID) based on the batch quantity and prints it as a QR code (or barcode) that can be affixed to the oil sample bottle or container. The system manages the generated codes uniformly and binds them to the database. The QR codes printed by the system can be randomly affixed to oil sample bottles or containers. Theoretically, oil sample bottles or containers with QR codes can be randomly used for oil sample collection (without absolute restrictions), but for ease of management, it is recommended to use them within the scope of the generated oil sample type and customer information.

[0018] S2. By scanning the identification code on the oil sample collection container with an authorized mobile terminal, the location information of the collection point, time information, collector's identity, collection purpose and oil sample collection container are bound and uploaded to the cloud platform to establish an oil source information database; Establish a database of oil source information, such as Figure 2 As shown, the authorized mobile terminal is equipped with an authorized APP or WeChat mini-program. The APP automatically locks and records the latitude, longitude, and time of the collection point. The collector completes the binding by scanning the identification code. The collector can supplement the information on the source and purpose of the oil sample as needed. The location information of the collection point includes latitude and longitude coordinates. Typically, before collecting oil samples, the collector will write down the collection point information and collection purpose on the label of the oil sample bottle or container as a basic label for oil sample identification. In this embodiment, the intelligent system enables the collector to use their smartphone to open an authorized APP (or WeChat mini program) to scan the QR code on the oil sample bottle or container instead of inputting the code.

[0019] S3. At the collection site, use a hyperspectral oil rapid detection device to perform spectral scanning on the collected oil sample to obtain spectral feature data. After associating the spectral feature data with the oil sample code, store it in the spectral feature database. At the collection site, a hyperspectral oil sample rapid detection device is used to generate the spectral characteristics of the collected oil sample in real time and store them in the spectral feature database along with the collection point information (corresponding to the oil sample code information). The hyperspectral oil sample rapid detection device is a portable device that does not require professional operation. In this embodiment, only this device is used to generate the spectral characteristics of the collected oil sample, which includes 300 bands. The spectral characteristics of the collected oil sample are prioritized for storage after the oil sample is bound to the collection sample. This can be understood as the fingerprint and permanent specimen of the collected oil sample being locked at the collection site, and together with the QR code, it becomes the index of the collected oil sample. like Figure 3 As shown, establishing a spectral feature database includes: The collector extracts 3.4 mL of oil sample from the oil sample container, inserts it into a cuvette, and then inserts the cuvette into a high-spectral oil rapid detection device. The device generates the spectral characteristics of the collected oil sample in real time. The spectral characteristics consist of 300 bands. At the same time, it can obtain the quantitative analysis components of the oil sample in real time, including metal components (such as iron, copper, cadmium) and other physicochemical indicators (such as concentration and moisture). The spectral characteristic data is preferentially stored in the spectral characteristic database along with the collection point information (corresponding to the oil sample code information).

[0020] S4. When the oil sample arrives at the laboratory and is stored, the laboratory testing personnel use a smart mobile terminal to scan the QR code on the container of the oil sample to bind the oil sample to the testing equipment, testing time, and testing steps, and directly import the equipment test results into the database to complete the oil sample storage registration. The process of oil samples arriving at the laboratory and being stored also includes: scanning or inputting paper test reports; when the oil sample being tested is selected as the modeling oil sample for the hyperspectral rapid oil detection equipment, the laboratory test results are input into the modeling system of the hyperspectral rapid oil detection equipment as the dependent variable of the hyperspectral feature model, updating the spectral quantitative analysis model, and realizing the quantitative analysis of this type of oil sample collected on-site by the hyperspectral rapid oil detection equipment through sample accumulation; The process of oil samples arriving at the laboratory for storage also includes oil sample aliquoting and testing. The aliquoting code of the oil sample sent for testing is linked to the oil sample code entering the storage: When the oil sample (sample bottle or container) arrives at the testing equipment, laboratory testing personnel can scan the QR code on the oil sample container using a special barcode scanner provided by the intelligent system, or a mobile terminal APP or WeChat mini program. This binds the oil sample to the testing equipment, testing steps, and time in the system. The equipment's test results can be directly imported into the database. The hyperspectral rapid oil detection equipment can perform quantitative analysis on oil samples collected on-site for this type of oil.

[0021] S5. Oil Sample Traceability: Based on the oil sample code and spectral characteristic data, record and trace the entire process of oil sample from collection, storage, testing, production process to delivery. Oil sample traceability includes two modes: consumer traceability and regulatory verification; The consumer traceability specifically includes: consumers scanning the traceability code affixed to the oil container with a mobile terminal to obtain information including place of origin, key indicators of production process, batch number, production time, and laboratory test report; The regulatory verification specifically includes: regulatory personnel extracting oil samples from oil containers, using high-spectral oil rapid detection equipment to conduct on-site real-time detection to obtain spectral characteristics, quantitatively monitoring the oil sample composition, tracing back to the batch number at the factory, and identifying the type of oil.

[0022] S6. Input the data from the oil source information database and the spectral feature database into the AI ​​multimodal model for intelligent analysis, predict, assess, warn, and evaluate the quality, safety, mixing degree, and safety risks of the oil sample, and generate an intelligent assessment report; By obtaining oil sample codes and spectral characteristics, the intelligent system retrieves and summarizes information on the origin, type, composition, production process, laboratory test indicators, manufacturer, and batch number of sampled vegetable oil samples in real time. Based on the application scenario, this information is directly pushed to the large model as a guide for data fusion to obtain targeted and intelligent solutions, including the degree of oil mixing, safety prediction, early warning, key component analysis, and solutions. AI-powered multimodal model-based intelligent analysis includes blended oil ratio estimation: An oil database collects the fatty acid composition and quality indicators of each raw oil, and a mathematical model is used to deduce the estimated proportions of each component in the blended oil. The specific process includes: The measured fatty acid composition vector of the blended oil is , Indicates the type and quantity of fatty acids (such as oleic acid, linoleic acid, etc.); The various raw material oils involved in the blending The fatty acid composition vector is ; The required solution is the ratio of the mixed oil. And satisfy ; Establish a mathematical model: ; when (Number of variables = number of equations + 1), which can be solved directly using a system of linear equations (least squares or Gaussian elimination). when (Undetermined, multiple solutions) It is necessary to introduce constrained optimization (such as least squares, non-negative least squares NNLS) to find the non-negative solution that is closest to the measured value; when (Overdetermined, redundant) The least squares method can be used to find the optimal approximate solution and evaluate the residuals; Based on the algorithm model, given the fatty acid composition of each raw material oil, the raw material oils and their proportions can be estimated by using a system of linear equations or the least squares method to estimate the composition of the tested food oil mixture. Safety risk prediction, assessment, early warning, and health assessment specifically include: AI multimodal model intelligent analysis of oil safety risk prediction, assessment and early warning and health assessment specifically includes: inputting oil traceability information, oil sample composition list images and consumer input information as multimodal data into a large language model, and extracting structured feature data, including acid value, peroxide value, brand, place of origin and batch; The structured feature data is input into the XGBoost gradient boosting decision tree model to obtain the risk score and risk level; The SHAP algorithm was used to calculate the marginal contribution of each feature in the risk score and to perform attribution analysis. The risk score, risk level, and SHAP attribution results are input into the large language model to generate an assessment report that includes natural language explanations, quantitative data, and traceability evidence.

[0023] This embodiment obtains a data system by constructing a digital vegetable oil industry chain. It uses a high-spectral oil rapid detection device to obtain spectral characteristics on-site and integrates the data system across the entire vegetable oil industry chain. This integrates the diverse data structure of vegetable oils and constructs a collaborative intelligent system of "information input - algorithm model - expert system". The intelligent system provides unstructured "data" as the starting point for "question and answer" or as an input tool for AI algorithm models, and has the functions of data integration and docking with other system modules, serving all aspects of vegetable oil production, factory exit, circulation and supervision.

[0024] Example 2: This embodiment provides a multi-source data analysis, tracing, and early warning system for plant oils, including: The oil sample coding generation module is used to generate a unique code for an oil sample and output it in the form of a QR code or barcode. The mobile terminal data acquisition module is used to scan the QR code and bind the data acquisition point information, time, collector, and acquisition purpose with the oil sample code. The hyperspectral detection module is used to acquire spectral characteristic data of oil samples in real time on site. The cloud platform database module is used to store the oil source information database and the spectral feature database, and to establish the relationship between the two; The laboratory testing data import module is used to import test results, equipment information, and paper reports into the database. The traceability module enables full-process tracing based on oil sample codes and spectral characteristics. The AI ​​multimodal analysis module is used to estimate the proportion of mixed oils and to predict, assess, and provide early warnings about the quality, safety, mixing degree, and safety risks of oil samples, generating intelligent assessment reports.

[0025] Example 3: The intelligent system generates a unique sample ID (UID) based on the collection requirements and prints a QR code, which is then affixed to the oil sample container. The intelligent system assists collection personnel (provided via an app or H5) by using their smartphones to scan the QR code on the oil sample container at the collection site to bind the oil sample and collection point. Clicking the scan icon initiates the scanning process, obtaining basic collection point information. Collection personnel can then input or supplement this information as needed, such as province, city, district, street, restaurant name, and season when collecting waste oil. The intelligent system automatically obtains the collection point's latitude and longitude, the collector's information, and the time of collection via the smartphone.

[0026] like Figure 5 As shown, on-site personnel can extract 3.4 mL of oil sample from the oil sample container, insert it into a cuvette, and then insert the cuvette into the hyperspectral oil rapid detection device to obtain the quantitative analysis components of the oil sample in real time, such as metal components (iron, copper, cadmium) and other physicochemical indicators (concentration, moisture); at the same time, the spectral characteristics are stored in the intelligent system spectral characteristic database.

[0027] The wind turbine gearbox lubricating oil whose spectral characteristics are collected on-site is recycled and reprocessed. When it is sent to the refinery, it can be sampled and tested using a high-spectral oil rapid detection device to obtain spectral characteristics and conduct real-time traceability verification. This can effectively avoid problems such as cross-contamination, mixing, and pollution during transportation from the wind turbine gearbox to the refinery.

[0028] When oil samples are delivered to the laboratory, the intelligent system assists laboratory personnel in scanning codes for storage and necessary sample sorting and testing. For example, the stored oil samples are sorted into different oil sample containers and sent to the laboratory for testing. The code on the oil sample container must be bound to the code on the stored oil sample container. This binding can be easily completed by scanning the code with a barcode scanner supported by the intelligent system. At each experimental stage, the experimental process is recorded by scanning codes, and experimental test reports, experimental equipment, and operating procedures are imported into the oil database.

[0029] Using the same method, the origin and processing technology of the oil can be entered into the oil database. When the oil leaves the factory, a QR code is generated by the information retrieval link in the oil database for that batch of oil. The oil leaving the factory can be batch-by-batch spectral features generated by a high-spectral oil rapid detection device and stored in the spectral feature library.

[0030] When oil products circulate in the market, consumers can scan the QR code printed with a search link on the oil container. The intelligent system will then display the oil's information from the database to the consumer's smartphone. In addition to providing information such as the oil product, production date, and place of origin, the traceability software allows consumers to click on the test report to learn more about the oil's composition and laboratory test reports. More information can be obtained through AI multimodal models.

[0031] Regulatory authorities can extract oil samples from the aforementioned oil containers and use hyperspectral rapid oil detection equipment for traceability verification; the spectral characteristics can be traced back to all detection stages (spectral characteristics are collected using hyperspectral rapid oil detection equipment), and although the tested oil samples may deteriorate or become ineffective over time, the spectral characteristics remain unchanged.

[0032] Example 4: Consumers are increasingly focusing on the safety indicators corresponding to oil quality and the health information associated with oil components during the distribution process.

[0033] like Figure 6 As shown, the intelligent system fully utilizes Alibaba's Qwen3.5 large model as the driving force for data fusion. Qwen3.5 adopts a native multimodal architecture, training text, images, videos, and audio in the same framework from the pre-training stage, so that visual perception and language reasoning share the same latent space from the beginning. It uses XGBoost (eXtremeGradient Boosting), a gradient boosting decision tree algorithm library, to calculate the average marginal contribution of each feature under all possible feature combinations in the overall multimodal model by weighted summation SHAP (SHapley Additive ex Planations) of the prediction results. Based on the traceability information of food oil and the oil information input by consumers, as well as related images, the intelligent system indirectly provides integrated regulations, standards, scientific research results, laboratory results, application cases, and relevant literature. Through the docking with the Qwen3.5 large model, it can provide the range of oil quality and safety indicators and information on the long-term health impact of use.

[0034] The practical operation process of multi-source data analysis, early warning, and health assessment methods for edible oils involves the following steps in the edible oil market circulation process: Consumers can trace the origin of edible oils by scanning the barcodes on the shelves and taking photos of the oil sample's component list, including brand and origin (brand = tea seed oil, origin = Liuyang, Hunan, batch = 240507). The intelligent system uploads oil traceability information, images of the oil circulating in the market, images of the oil sample's component list (unstructured data), and common food safety and health-related questions. Through the multimodal model Qwen3.5, structural characteristics (acid value = 0.4) are obtained. (Peroxide value = 0.18, Brand = Camellia seed oil, Origin = Liuyang, Hunan, Batch = 240507); The data was further processed into the XGBoot model to obtain numerical predictions: Risk score 78.3 points + Level III; Attribution analysis was performed using the SHAP algorithm: Acid value contributed 32%, and origin contributed 28%; Finally, a report was generated using the multimodal model Qwen3.5: Integrating XGBoost score + SHAP attribution; Real-time output: An assessment report with natural language + quantitative data + traceability evidence, including statistical data on health assessment.

[0035] Example 5: This embodiment illustrates the process of tracing the origin of mixed oils: Laboratory measurements of camellia seed oil samples showed 76% oleic acid and 9.5% linoleic acid, indicating that the oil sample had been mixed. However, the experimental methods could not determine the composition and proportion of the mixed raw materials. Assuming the intelligent system database contains three types of camellia seed oil (A, B, and C), with oleic acid and linoleic acid ratios as shown in Table 1: Table 1

[0036] Based on the algorithm model, the equation is established, and its expression is: ; ; .

[0037] Due to detection errors, the fatty acid content of camellia seed oil itself has natural fluctuations (not a fixed constant), uneven mixing or sampling deviations, etc., in practice, an exact solution is not required, but a least squares approximate solution is sought, that is, the sum of squares of the deviations between the predicted value and the measured value is minimized. because The objective function becomes only containing and The expression for a quadratic function in two variables is: ; After simplification, and The quadratic function opens upwards. Taking the partial derivative, we get the unconstrained optimal solution. We then check if it satisfies ≥0. If not, we re-solve using boundary constraints (setting a variable to 0) to obtain the constrained least squares solution. (That is, the proportion of camellia seed oil A is 50%) (That is, the proportion of camellia seed oil B is 30%) (That is, the proportion of C in camellia seed oil is 20%) The laboratory, relying on an intelligent system, not only obtained the test data but also provided a prediction or assessment of the degree of blending of the camellia seed oil.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for multi-source data analysis, tracing, and early warning of plant oils, characterized in that, Includes the following steps: S1. Oil Sample Code Generation: Generate a unique code UID for the oil sample based on the oil sample type, customer type, collection type, and collection batch quantity. Output the code as a QR code or barcode and affix it to the oil sample container. The oil sample code is bound to the database. S2. By scanning the identification code on the oil sample collection container with an authorized mobile terminal, the location information of the collection point, time information, collector's identity, collection purpose and oil sample collection container are bound and uploaded to the cloud platform to establish an oil source information database; S3. At the collection site, use a hyperspectral oil rapid detection device to perform spectral scanning on the collected oil sample to obtain spectral feature data. After associating the spectral feature data with the oil sample code, store it in the spectral feature database. S4. When the oil sample arrives at the laboratory and is stored, the identification code on the oil sample container is scanned and collected by a smart mobile terminal to register the oil sample for storage. S5. Oil Sample Traceability: Based on the oil sample code and spectral characteristic data, record and trace the entire process of oil sample from collection, storage, testing, production process to delivery. S6. Input the data from the oil source information database and the spectral feature database into the AI ​​multimodal model for intelligent analysis, predict, assess, warn, and evaluate the quality, safety, mixing degree, and safety risks of the oil sample, and generate an intelligent assessment report.

2. The method for multi-source data analysis, tracing, and early warning of vegetable oils according to claim 1, characterized in that, The authorized mobile terminal is equipped with an authorized APP or WeChat mini program. The collector completes the binding by scanning the identification code. The collector can supplement the source information of the collected oil sample as needed. The location information of the collection point includes latitude and longitude coordinates.

3. The method for multi-source data analysis, tracing, and early warning of vegetable oils according to claim 1, characterized in that, Laboratory testing personnel use smart mobile terminals to scan the QR code on the container of the oil sample being tested, linking the oil sample with the testing equipment, testing time, and testing steps, and directly importing the equipment test results into the database. The process of oil samples arriving at the laboratory for storage also includes: scanning or inputting paper test reports; when the tested oil sample is selected as the modeling oil sample for the hyperspectral rapid oil detection equipment, the laboratory test results are input into the modeling system of the hyperspectral rapid oil detection equipment as the dependent variable of the hyperspectral feature model, updating the spectral quantitative analysis model, and realizing the quantitative analysis of similar oil samples collected on-site by the hyperspectral rapid oil detection equipment through sample accumulation.

4. The method for multi-source data analysis, tracing, and early warning of vegetable oils according to claim 3, characterized in that, The process of oil samples arriving at the laboratory for storage also includes oil sample sub-samples for testing, and the sub-sample code at the time of sub-sample testing is linked to the oil sample code upon entering the storage.

5. The method for multi-source data analysis, tracing, and early warning of vegetable oils according to claim 1, characterized in that, Oil sample traceability includes two modes: consumer traceability and regulatory verification; The consumer traceability specifically includes: consumers scanning the traceability code affixed to the oil container with a mobile terminal to obtain information including place of origin, key indicators of production process, batch number, production time, and laboratory test report; The regulatory verification specifically includes: regulatory personnel extracting oil samples from oil containers, using high-spectral oil rapid detection equipment to conduct on-site real-time detection to obtain spectral characteristics, quantitatively monitoring the oil sample composition, tracing back to the batch number at the factory, and identifying the type of oil.

6. The method for multi-source data analysis, tracing, and early warning of vegetable oils according to claim 1, characterized in that, The AI ​​multimodal model intelligent analysis includes blended oil ratio estimation: the oil database collects the fatty acid composition and quality indicators of each raw oil, and uses mathematical models to deduce the estimated ratio of each component in the blended oil.

7. The method for multi-source data analysis, tracing, and early warning of vegetable oils according to claim 6, characterized in that, The AI ​​multimodal model intelligent analysis of oil safety risk prediction, assessment, early warning and health assessment specifically includes: inputting oil traceability information, oil sample component list images and consumer input information as AI multimodal data into the big language model, and extracting structured feature data, including acid value, peroxide value, brand, place of origin and batch; The structured feature data is input into the XGBoost gradient boosting decision tree model to obtain the risk score and risk level; The SHAP algorithm was used to calculate the marginal contribution of each feature in the risk score and to perform attribution analysis. The risk score, risk level, and SHAP attribution results are input into the large language model to generate an assessment report that includes natural language explanations, quantitative data, and traceability evidence.

8. A multi-source data analysis, traceability, and early warning system for vegetable oils, employing the method described in any one of claims 1-7, characterized in that, include: The oil sample coding generation module is used to generate a unique code for an oil sample and output it in the form of a QR code or barcode. The mobile terminal data acquisition module is used to scan the QR code and bind the data acquisition point information, time, collector, and acquisition purpose with the oil sample code. The hyperspectral detection module is used to acquire spectral characteristic data of oil samples in real time on site. The cloud platform database module is used to store the oil source information database and the spectral feature database, and to establish the relationship between the two; The laboratory testing data import module is used to import test results, equipment information, and paper reports into the database. The traceability module enables full-process tracing based on oil sample codes and spectral characteristics. The AI ​​multimodal analysis module is used to estimate the proportion of mixed oils and to predict, assess, and provide early warnings about the quality, safety, mixing degree, and safety risks of oil samples, generating intelligent assessment reports.