Agricultural product circulation traceability method and system

Abstract

The present invention relates to the field of agricultural product circulation management and traceability technology, and in particular to an agricultural product circulation traceability method and system, the method comprising: recording temperature, humidity, light and other environmental conditions in real time through smart material tags, and dynamically binding to digital twins; combining improved isolation forest algorithm to detect environmental anomalies and generate anomaly detection results; using recursive neural network model and multivariate interpolation algorithm to predict trends of completed time series data and generate future state data; constructing a knowledge graph containing historical, current and future states based on digital twins and predicted data, and generating traceability paths and risk warnings. The present invention realizes real-time monitoring, accurate prediction and intelligent traceability of the agricultural product circulation process, improves risk warning capabilities and data transparency, and provides reliable decision support for consumers, producers and regulators.

Description

Technical Field

[0001] The present invention relates to the field of agricultural product circulation management and traceability technology, and in particular to an agricultural product circulation traceability method and system. Background Art

[0002] Agricultural product traceability is of great significance in modern agriculture. It can provide consumers with transparent information on product sources, production conditions, and transportation environments, enhance consumer confidence, and provide regulators with accurate quality monitoring tools. However, due to the environmental complexity of agricultural products during circulation (such as changes in temperature, humidity, light, etc.) and the dynamic nature of transportation links, real-time monitoring and anomaly detection face technical challenges. An efficient traceability system requires not only the ability to record historical information, but also the ability to predict future environmental conditions to achieve intelligent decision support.

[0003] The existing technology adopts the data collection, verification and classification management method (China invention patent, publication number: CN118152633A, name: Data verification management system and method for agricultural information digital management platform) to collect agricultural information through satellite remote sensing technology, IoT sensor equipment and high-resolution monitoring equipment, and verify and classify the data. However, these technologies have the following shortcomings:

[0004] Existing solutions mainly focus on the collection and processing of static data, and lack the ability to monitor the dynamic environmental status during the circulation process in real time; existing solutions rely on predefined verification rules to identify abnormal data, and fail to adaptively adjust according to the dynamic changes of actual data, resulting in decreased detection accuracy in complex scenarios; existing solutions fail to predict status based on circulation data, and it is difficult to provide early warning and optimization suggestions for potential risks; existing solutions only perform static classification in data management, and do not conduct correlation analysis between the status of agricultural products and environmental information, which limits the depth of data application. Summary of the invention

[0005] In view of the many problems existing in the above-mentioned prior art, the present invention provides a method and system for tracing the circulation of agricultural products. The present invention is based on digital twin construction, dynamic monitoring and state prediction, records environmental data through smart material tags, dynamically binds real-time states to digital twins, and combines improved isolation forest algorithm and recursive neural network model for anomaly detection and trend prediction. At the same time, by building a knowledge graph to associate historical, current and future states, a traceability path containing environmental changes and potential risk prompts is generated, which improves the transparency and intelligence level of agricultural product circulation management.

[0006] A method for tracing the origin of agricultural products, comprising the following steps:

[0007] Obtaining RGB image data, environmental data, and geographic location information of the agricultural product, and integrating all the data to generate a digital twin, wherein the digital twin includes a unique identifier of the agricultural product, a recorded environmental status, and a geographic location;

[0008] In the circulation of agricultural products, smart material tags are used to record changes in temperature, humidity, light and other environmental conditions in real time, and the state data recorded by the tags are dynamically bound to the digital twin. Anomaly detection is performed on the dynamically bound data, and anomalies in the environmental state are identified by combining the improved isolation forest algorithm to generate anomaly detection results.

[0009] Based on the digital twin and anomaly detection results, data association analysis is performed, and the reconstructed digital twin is generated by optimizing the structure of the digital twin. A recursive neural network model combined with a multivariate interpolation algorithm is used to complete the anomaly points and predict the future state to generate predicted state data.

[0010] Extract the reconstructed digital twin data and predicted status data, and build a knowledge graph that includes the status of agricultural products, circulation paths, and future predicted status; generate a traceability path based on the knowledge graph that includes historical status, current status, and future risk warnings.

[0011] Preferably, the step of generating a digital twin comprises:

[0012] The RGB image data of agricultural products is obtained through RGB imaging equipment, and the edge features in the RGB image data are processed using the morphological gradient enhancement algorithm to highlight the texture features of leaf veins or fruit surfaces. The processed texture feature data is embedded in combination with the deep contrast learning model to generate the biological texture identification data of each agricultural product, and the biological texture identification data is integrated with the environmental data and geographic location information to generate a digital twin.

[0013] Preferably, the environmental data includes temperature and humidity data and gas concentration data, the temperature and humidity data are collected by temperature and humidity sensors, and the gas concentration data are obtained by detecting the concentration level of ethylene gas; the geographic location information is obtained by high-precision GNSS equipment, and the data in weak signal areas are corrected based on a dynamic geographic tagging algorithm.

[0014] Preferably, the smart material label comprises a thermochromic particle film and a spectrally responsive coating, wherein the thermochromic particle film changes color within a specific threshold range according to changes in ambient temperature, and the spectrally responsive coating reflects light signals of specific wavelengths under different lighting conditions; the color changes and spectral reflectance characteristics of the label are recorded by a high-resolution image acquisition device, and real-time label status data is generated.

[0015] Preferably, the step of performing anomaly detection on dynamically bound data includes:

[0016] Combined with the improved isolation forest algorithm, by constructing a random partition tree of multidimensional data, the temperature and humidity values, texture feature values ​​and time series recorded in the label status data are segmented layer by layer; the distribution density and the degree of outlier of data points in each segmented area are analyzed, anomalies are identified and anomaly detection results are generated, which include the occurrence time, impact range and outlier score of the anomaly.

[0017] Preferably, the improved isolation forest algorithm adopts a dynamic adjustment strategy for the random partitioning of label status data, sets the partition depth and the number of partition trees according to the non-uniformity of data distribution, and determines the credibility of outliers by calculating the cumulative distribution probability of outliers.

[0018] Preferably, the step of reconstructing the digital twin includes:

[0019] By adding associated dimensions to the structure of the digital twin, including the spectral change characteristics of the environmental state and the temperature and humidity change trends; combining the multivariate interpolation algorithm to complete the data near the abnormal points, and using the Lagrange interpolation method to smooth the values ​​of the continuous time series, the completed time series data is generated, and the digital twin is reconstructed based on this.

[0020] Preferably, the step of predicting the future state includes:

[0021] The recursive neural network model is used to learn the completed time series data and build a prediction model. By inputting a continuous environmental state time series, the predicted state data for the future time period is generated. The predicted state data includes the temperature and humidity trends of the future environment, changes in spectral characteristics, and the location and impact range of potential abnormal points.

[0022] Preferably, the construction of the knowledge graph includes:

[0023] Nodes and associated paths containing agricultural product status information are extracted from the reconstructed digital twin data, and nodes are defined as status types or location information, and paths are defined as associations between data. Combined with the predicted status data, the future status is mapped into virtual nodes in the knowledge graph, and a knowledge graph containing historical status, current status, and future status association paths is generated, which is converted into structured data based on the resource description framework for storage and query.

[0024] An agricultural product circulation traceability system, used to implement the agricultural product circulation traceability method, comprises:

[0025] A data acquisition module, used to obtain RGB image data, environmental data and geographic location information of agricultural products, the data acquisition module includes an RGB imaging device, a temperature and humidity sensor, a gas sensor and a high-precision geographic positioning device to collect texture feature data, temperature and humidity data, gas concentration data and geographic tag data of agricultural products;

[0026] A digital twin generation module is used to integrate the RGB image data, environmental data and geographic tag data obtained by the data acquisition module to generate a digital twin containing a unique identifier, environmental status and geographic location;

[0027] A state recording module, including a smart material tag and a state acquisition device. The smart material tag is used to record temperature, humidity, light changes and other environmental conditions in real time. The state acquisition device is used to extract state data from the smart material tag and dynamically bind it to the digital twin.

[0028] An anomaly detection module is used to detect anomalies in the environment state based on the dynamic binding data generated by the state recording module in combination with the improved isolation forest algorithm, and generate anomaly detection results, wherein the anomaly detection results include the occurrence time, dimension and outlier score of the anomaly;

[0029] The twin reconstruction and prediction module is used to optimize the structure of the digital twin according to the anomaly detection results to generate the reconstructed digital twin, and to complete the data near the anomaly point by combining the multivariate interpolation algorithm, and to analyze the time series data through the recursive neural network model to generate the predicted status data;

[0030] The knowledge graph construction module is used to extract key nodes and associated paths from the reconstructed digital twin data and predicted state data, construct a knowledge graph containing historical states, current states, and future predicted states, and generate traceability path data;

[0031] The feedback generation module is used to generate customized feedback reports for consumers, producers and regulatory agencies based on the knowledge graph. The feedback reports include the traceability path of agricultural products, status history records and future risk warnings.

[0032] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0033] The present invention realizes real-time monitoring of environmental conditions such as temperature, humidity, and light during the circulation process through the dynamic binding technology of smart material tags and digital twins;

[0034] The present invention improves the accuracy and robustness of anomaly detection by combining an improved isolation forest algorithm with a dynamic partitioning strategy and cumulative distribution probability calculation.

[0035] The present invention uses a recursive neural network model and a multivariate interpolation algorithm to achieve accurate prediction of future environmental conditions during the circulation process, providing a reliable basis for risk warning and optimized decision-making;

[0036] The present invention provides visual information display and intelligent query support for the entire life cycle of agricultural products by constructing a knowledge graph that includes historical, current and future states, generating traceability paths and risk warnings;

[0037] The present invention realizes efficient integration and dynamic updating of data by adding new correlation dimensions and dynamically optimizing the twin structure, providing comprehensive support for anomaly analysis and tracing. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the process of the present invention;

[0039] Figure 2 A schematic diagram of the anomaly detection and result generation process in the present invention;

[0040] Figure 3 This is a schematic diagram of the future state prediction process in the present invention;

[0041] Figure 4 A schematic diagram for knowledge graph construction and traceability path generation in the present invention;

[0042] Figure 5 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0043] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0044] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0045] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0046] like Figure 1 As shown, a method for tracing the origin of agricultural products includes the following steps:

[0047] Obtaining RGB image data, environmental data, and geographic location information of the agricultural product, and integrating all the data to generate a digital twin, wherein the digital twin includes a unique identifier of the agricultural product, a recorded environmental status, and a geographic location;

[0048] RGB image data is mainly used to record the appearance characteristics of agricultural products, including color, shape, surface texture, etc. Through high-resolution RGB imaging devices (such as industrial cameras and smartphone cameras), agricultural products can be photographed in a standardized manner, and key features can be extracted in combination with image processing algorithms for individual identification and quality assessment.

[0049] Use a high-resolution camera (≥12 million pixels) to capture images of agricultural products to ensure sufficient detail resolution. Use standardized light sources (such as LED uniform lighting) to eliminate shadows and light interference and ensure image consistency. Use morphological gradient enhancement algorithms to extract edge features to highlight the texture details of leaf veins or fruit surfaces and improve individual recognition accuracy. Use local contrast enhancement processing to improve the distinguishability of surface details of different agricultural products. Use a deep contrast learning model to extract image features and generate biotexture identification data for unique identification of agricultural products.

[0050] Environmental data is a record of the environmental conditions of agricultural products during circulation, including temperature and humidity data, light intensity, gas concentration, etc. These data are collected in real time through sensors and serve as an important dimension of the digital twin to ensure that the storage and transportation conditions of agricultural products are traceable.

[0051] The ambient temperature and humidity during the circulation process are collected through the temperature and humidity sensor. The data format is:

[0052]

[0053] in, is the timestamp, is the temperature value, is the humidity value.

[0054] Light sensors are used to detect light levels at different time periods and obtain light intensity data to ensure that agricultural products are stored under appropriate light conditions.

[0055] The ethylene concentration (gas concentration data) is detected by gas sensors to evaluate the maturity of fruits and vegetables, and multi-dimensional analysis is performed in combination with RGB image features.

[0056] Geographic location information is used to mark the circulation path of agricultural products, including picking points, storage points, transportation routes, etc. Geographic coordinates are obtained through high-precision GNSS equipment, and errors are corrected based on dynamic geographic tagging algorithms to ensure the accuracy of location information.

[0057] Dual-frequency GNSS equipment is used to record the geographic coordinates of agricultural products when they are picked, and the reference station signal is combined to improve positioning accuracy. In areas where signals are interfered with (such as areas blocked by buildings), fuzzy logic inference algorithms are used to correct positioning data to ensure data integrity. The format of the recorded geographic location information is as follows:

[0058]

[0059] in, , are latitude and longitude respectively, Is the timestamp.

[0060] The digital twin is the unique mapping of agricultural products in virtual space. By integrating RGB image data, environmental data and geographic information, it generates a digital model that can be dynamically updated to support agricultural product traceability, anomaly detection and future status prediction.

[0061] Combined with deep learning feature encoding of RGB image data, a biotexture identifier is generated for each agricultural product to ensure individual distinguishability. Environmental data such as temperature, humidity, and light intensity are synchronized with the corresponding timestamp to record environmental conditions during the circulation process. Using the timestamp as the key, the geographic location information is bound to the digital twin to achieve location tracking. Digital twin structure:

[0062]

[0063] in, is the biological texture identification data, is the environmental state, For geographic location information, It is a unified timestamp for the above data.

[0064] Through RGB image feature extraction and deep contrast learning, individual-level identification of agricultural products is achieved, enhancing the accuracy of the traceability system. The integration of environmental data such as temperature, humidity, light, and gas concentration provides a high-precision environmental description for the status monitoring of agricultural products during the circulation process. Combined with high-precision GNSS equipment and dynamic geographic tagging algorithms, the geographical location information of agricultural products in the circulation link is accurate, improving the transparency of traceability. The establishment of digital twins provides complete data support for subsequent anomaly detection, future status prediction, and knowledge graph construction, realizing intelligent agricultural product traceability management.

[0065] Through the feature encoding of RGB image data, unique identification of individual agricultural products is achieved, which greatly improves the accuracy of traceability; by integrating environmental data and geographic location information, the environmental conditions and spatial position at the time of collection are comprehensively recorded, providing data support for subsequent anomaly detection and traceability analysis; the digital twin generated by deep learning and data fusion algorithms has high stability and scalability, and can be applied to subsequent circulation status monitoring and prediction; the construction of digital twins has laid a high-quality data foundation for the subsequent construction of knowledge graphs, and realized the deep integration of agricultural product status and spatial circulation information.

[0066] Preferably, the step of generating a digital twin comprises:

[0067] The RGB image data of agricultural products is obtained through RGB imaging equipment, and the edge features in the RGB image data are processed using the morphological gradient enhancement algorithm to highlight the texture features of leaf veins or fruit surfaces. The processed texture feature data is embedded in combination with the deep contrast learning model to generate the biological texture identification data of each agricultural product, and the biological texture identification data is integrated with the environmental data and geographic location information to generate a digital twin.

[0068] The present invention obtains RGB image data of agricultural products through high-resolution RGB imaging equipment, and processes edge features in combination with morphological gradient enhancement algorithm to highlight the texture features of leaf veins or fruit surfaces, thereby enhancing individual recognition capabilities. Subsequently, the processed texture feature data is feature embedded using a deep contrast learning model to generate biological texture identification data, which is then integrated with environmental data and geographic location information to construct a complete digital twin. Digital twins are not only used for unique identification of agricultural products, but also can dynamically update status information to ensure the accuracy and completeness of agricultural product circulation traceability.

[0069] RGB image data records key features of agricultural products such as color, morphology, and surface texture, and is used for individual identification and quality assessment. Compared with multispectral imaging, RGB imaging devices are more widely used, and can improve data availability through image enhancement and feature extraction methods. This solution uses high-resolution RGB cameras to perform standardized imaging of agricultural products, and uses morphological gradient enhancement algorithms to optimize edge features to ensure individual differentiation capabilities.

[0070] Use high-resolution RGB cameras with more than 12 megapixels to ensure that the surface texture of agricultural products is clearly visible. Eliminate shadows and improve shooting consistency through standardized light sources (such as LED uniform lighting).

[0071] The morphological gradient enhancement algorithm is used to calculate the local gradient value to highlight the edge features and enhance the surface details of agricultural products, such as the veins of leaves or micro cracks on the surface of fruits, to improve the accuracy of individual recognition. Through local contrast enhancement processing, the recognizability of different surface areas of agricultural products is improved to ensure effective segmentation under complex backgrounds. Gradient enhancement formula:

[0072]

[0073] in, is the gradient image, For the input RGB image, is the structural element, and Represent the dilation and erosion operations respectively. By choosing the appropriate size and shape of the structure element, the degree of enhancement of the texture features can be flexibly controlled.

[0074] Biotexture identification data is used to uniquely identify individual agricultural products. The deep contrast learning model learns the texture features of RGB images, maps them into high-dimensional feature vectors, and ensures the feature similarity of the same agricultural product individuals, while distinguishing different individuals to achieve accurate traceability.

[0075] ResNet or Vision Transformer (ViT) models are used to perform deep learning on texture features in RGB images to generate highly discriminative feature vectors.

[0076] The contrastive loss function is trained to optimize the feature representation, minimize the embedding distance of the same agricultural product samples, and maximize the embedding distance of different individuals.

[0077] Calculate the unique characteristic code of agricultural products and convert it into a unique identifier through a hash function to ensure traceability during circulation.

[0078] Contrastive loss formula:

[0079]

[0080] in, is the loss function, Indicates whether the sample pair is of the same type. is the Euclidean distance between feature vectors, is the threshold value.

[0081] By integrating bio-texture identification data, environmental data and geographic information, a digital twin containing multi-dimensional information is generated. The digital twin can be dynamically updated to reflect the status changes of agricultural products in real time, and support traceability, anomaly detection and status prediction.

[0082] The storage environment data of agricultural products is collected through temperature and humidity sensors and light sensors, and aligned with the RGB image data timestamp to ensure data consistency.

[0083] Recorded data format:

[0084]

[0085] in, is the temperature value, is the humidity value, is the light intensity, Is the timestamp.

[0086] High-precision GNSS equipment is used to record the geographic coordinates of agricultural products picked or stored, and dynamic geo-tagging algorithms are used for error correction.

[0087] The format of recorded geographic location information is:

[0088]

[0089] in, is the latitude and longitude, is the corresponding timestamp.

[0090] Digital twin data structure:

[0091]

[0092] in, is the biological texture identification data, is the environmental state, For geographic location information, It is a unified timestamp for the above data.

[0093] Through deep learning feature extraction of RGB images, the unique tracking of agricultural products is achieved to ensure the accuracy of traceability. Environmental data such as temperature, humidity, light, and gas concentration are integrated to provide dynamic monitoring of the entire life cycle and improve the comprehensiveness of data. Combined with high-precision GNSS positioning and dynamic geo-tagging algorithms, the location records of agricultural products in the supply chain are ensured to be accurate and traceability is improved. Digital twins achieve efficient organization and real-time updating of data, providing data support for subsequent anomaly detection, trend prediction, and intelligent analysis.

[0094] Example: An organic fruit supply chain company hopes to use digital twins to manage the storage and transportation of high-value fruits (such as imported blueberries) to ensure their quality traceability. Implementation process:

[0095] The surface texture of blueberries is photographed with a high-resolution RGB camera, and features are extracted using a deep learning model to generate bio-texture identification data. The temperature, humidity, and light intensity of the blueberry storage environment are recorded, and the storage location data is obtained based on the GNSS device. RGB image data, environmental data, and geographic information are integrated to build a digital twin of blueberries and assign a unique identifier. During storage and transportation, the temperature, humidity, and light change data are continuously updated and bound to the digital twin to ensure the integrity of the traceability data.

[0096] Logistics companies can accurately monitor the storage environment of blueberries and optimize transportation conditions; consumers can check the source and distribution process of products and enhance their trust in food safety.

[0097] Preferably, the environmental data includes temperature and humidity data and gas concentration data, the temperature and humidity data are collected by temperature and humidity sensors, and the gas concentration data are obtained by detecting the concentration level of ethylene gas; the geographic location information is obtained by high-precision GNSS equipment, and the data in weak signal areas are corrected based on a dynamic geographic tagging algorithm.

[0098] The temperature and humidity sensors are used to monitor the air temperature and humidity status at the agricultural product collection point in real time. The temperature and humidity data reflect the external environment when picking, which has a direct impact on the preservation quality of agricultural products. For example, a low temperature and high humidity environment may cause condensation on the surface of some fruits, increasing the risk of mold growth.

[0099] High-sensitivity temperature and humidity sensors are installed at the picking points to collect air temperature and relative humidity values ​​in real time. The collected data is recorded in the form of time series:

[0100]

[0101] in, is the timestamp, is the temperature value, is the humidity value. The sliding average algorithm is used to denoise the continuous data to ensure data stability and output the cleaned temperature and humidity data stream.

[0102] The gas concentration sensor detects the ethylene gas concentration level to assist in determining the maturity of agricultural products. Ethylene is a plant hormone, and its concentration changes are closely related to the respiration and ripening process of the fruit. Based on the working mechanism of the electrochemical sensor, the concentration of ethylene gas is detected through redox reactions. Sensors are arranged in the picking area to continuously monitor the ethylene concentration in the air and generate data records:

[0103] in, is the timestamp, is the ethylene concentration value. Combined with the known ethylene concentration standard sample, the sensor data is calibrated to ensure the accuracy of the measurement.

[0104] The longitude and latitude of the picking point are obtained through high-precision GNSS equipment. The geographic location information is used to mark the spatial location of agricultural product picking, providing a basis for subsequent circulation path analysis. A dual-frequency GNSS receiver is used to receive satellite signals to reduce the impact of multipath errors and ionospheric delay on positioning accuracy. An information stream containing longitude, latitude and timestamp is generated:

[0105]

[0106] in, is the latitude, is the longitude, Is the timestamp.

[0107] In weak signal areas (such as forests, valleys or densely built-up areas), GNSS signals may be interfered by multipath effects, resulting in positioning errors. The present invention adopts a dynamic geo-tagging algorithm to perform real-time correction on data in weak signal areas to improve the accuracy of geographic location information. The received multipath signals are weighted to remove abnormal signals, and preliminary correction data is generated based on weighted average:

[0108]

[0109] in, is the path weight, is the position value corresponding to the path signal.

[0110] Based on the known landmark locations and the area map, the actual location of the picking point is inferred through fuzzy rules. For example:

[0111] Rule 1: If the signal confidence is low and the location deviates from the regional landmark, adjust the longitude and latitude to get closer to the landmark.

[0112] Rule 2: If the confidence of multiple signals is similar, the centroid is taken as the correction value.

[0113] Output the corrected longitude and latitude values ​​and their confidence levels to generate high-precision geographic location information.

[0114] The environmental data obtained through temperature and humidity sensors and gas concentration sensors comprehensively reflect the external conditions during picking, providing accurate input for subsequent anomaly detection and status prediction; combined with high-precision GNSS equipment and dynamic geo-tagging algorithms, accurate acquisition of geographic location information is achieved, ensuring accurate positioning of picking points even in weak signal areas; environmental data and geographic location information, as key dimensions of digital twins, support transparent tracking of the entire circulation path of agricultural products, and provide a highly reliable data foundation for abnormal batch positioning and quality monitoring.

[0115] Embodiment: A batch of blueberries is picked on a mountain farm. To ensure the accuracy of environmental records in circulation traceability, it is necessary to collect temperature and humidity data, ethylene concentration, and geographic location information. Temperature and humidity sensors and ethylene concentration sensors are arranged in the picking area to collect temperature and humidity values ​​and ethylene concentration data in real time to generate environmental data records. A dual-frequency GNSS device is used to record the longitude and latitude information of the picking point. For areas with large signal interference, the positioning error is corrected by a dynamic geo-tagging algorithm. The collected environmental data and geographic location information are bound to the digital twin of each blueberry.

[0116] The digital twin of each blueberry fully records the temperature and humidity, ethylene concentration and accurate geographic location information at the time of picking, providing precise support for status monitoring and abnormal positioning during subsequent transportation, while also enhancing consumers' trust in the quality source of blueberries.

[0117] like Figure 2 As shown in the figure, in the circulation of agricultural products, smart material tags are used to record changes in temperature, humidity, light and other environmental conditions in real time, and the state data recorded by the tags are dynamically bound to the digital twin; anomaly detection is performed on the dynamically bound data, and the improved isolation forest algorithm is combined to identify abnormal points in the environmental state to generate anomaly detection results;

[0118] Smart material tags are designed based on thermochromic microparticle film and spectrally responsive coating technology, using the physical properties of materials under changing environmental conditions to record parameters such as temperature, humidity and light in real time. These tags are passive and can sense and record changes in environmental conditions without external energy support.

[0119] Thermochromic microparticle film changes color when the temperature changes through heat-sensitive materials. For example, when the temperature rises from below 5°C to 25°C, the label color gradually changes from blue to red. The color change value is captured by imaging equipment and quantified as a temperature value.

[0120] Spectrally responsive coatings reflect light signals of different wavelengths under different light intensities. High-resolution spectral sensors are used to capture the reflected signals and generate light intensity data.

[0121] Each smart material tag is bound to the digital twin, and data association is achieved through a unique identifier. The tag surface is scanned regularly using an image acquisition device (such as a camera) to capture changes in color and spectrum and generate tag status data. The tag status data is uploaded to the central data processing unit in real time through a wireless transmission device and dynamically bound to the corresponding digital twin.

[0122] Dynamic binding is the process of associating the tag status data collected during the circulation process with the digital twin of agricultural products. Through binding, the digital twin can update the circulation status of agricultural products in real time, reflecting the changes in environmental conditions during transportation and storage. Based on the unique identifier of the tag, the tag status data is matched with the initial information in the digital twin (such as texture features and environmental data of the collection point). With the timestamp as the primary key, the dynamic tag status data is appended to the state dimension of the digital twin to form a time series.

[0123] The improved isolation forest algorithm is used to analyze the dynamically bound tag status data to detect abnormal points in the circulation process. The improved isolation forest algorithm is based on a random partition tree, which analyzes the distribution density and outlier degree of the data and can effectively identify abnormal conditions, such as excessive temperature and humidity or abnormal light.

[0124] By randomly selecting data dimensions and split points, a partition tree is constructed. The path length of each data point reflects its distribution density, and outliers have shorter path lengths. The anomaly score of each data point is calculated based on the partition tree. The higher the anomaly score, the more likely the data point is an anomaly. Output: Generate anomaly detection results, including the time, dimension, and score of the anomaly point.

[0125] Preferably, the smart material label comprises a thermochromic particle film and a spectrally responsive coating, wherein the thermochromic particle film changes color within a specific threshold range according to changes in ambient temperature, and the spectrally responsive coating reflects light signals of specific wavelengths under different lighting conditions; the color changes and spectral reflectance characteristics of the label are recorded by a high-resolution image acquisition device, and real-time label status data is generated.

[0126] Thermochromic microparticle film is a smart material based on thermosensitive materials, whose molecular structure changes within a specific temperature threshold, thereby changing its optical properties (such as color). This change is reversible, that is, when the temperature returns to the normal range, the color of the film will return to its original state.

[0127] When the temperature reaches the activation point of the film, the molecular structure rearranges, causing the light reflecting properties to change, thus changing the color of the label. For example:

[0128] When the temperature is below 5℃, the label is blue; when the temperature rises to 25℃, the label gradually turns red.

[0129] According to the storage conditions of agricultural products, thermochromic films can be designed with different threshold ranges. For example, the temperature sensitivity range of cold chain fruits is 0℃ to 10℃, and the color change range corresponding to the thermochromic film can be set to this temperature range. The label color is captured in real time by a high-resolution image acquisition device, and its RGB value is extracted and quantified into a temperature value:

[0130]

[0131] in, is the temperature value, are the red, green, and blue component values ​​of the label color, is the temperature mapping function, obtained through experimental calibration.

[0132] A spectrally responsive coating is a material that reflects light signals of a specific wavelength under specific lighting conditions. By monitoring the spectral characteristics of the coating's reflection, the intensity of ambient light can be quantified. When the light intensity changes, the coating's reflection intensity for a specific wavelength also changes. For example:

[0133] Under low light conditions (<200 lux), the coating mainly reflects long waves (infrared); under strong light conditions (>1000 lux), the short waves (blue light) reflected by the coating dominate. The reflected light signal of the coating is captured by a high-resolution spectral sensor to generate spectral data:

[0134]

[0135] in, is the light intensity value, is the wavelength, For light intensity, is the mapping function from spectrum to intensity.

[0136] To adapt to the various lighting environments of agricultural product storage, the spectral response range of the coating usually covers 400-700 nm (visible light) and 700-1000 nm (near infrared).

[0137] The color change of the thermochromic particle film and the spectral characteristics of the spectral response coating are captured in real time through image acquisition equipment to generate tag status data reflecting the environmental status. Equipped with high-resolution imaging equipment with a resolution of more than 20 million pixels, it supports real-time shooting and data transmission. The results of each acquisition are recorded as time series data, including timestamp, temperature value and light intensity value:

[0138]

[0139] in, is the timestamp, is the temperature value, is the light intensity value.

[0140] Smart material tags record temperature and light changes in real time to ensure that environmental conditions during circulation can be tracked throughout the entire process; thermochromic particle films and spectrally responsive coatings are sensitive to changes in environmental conditions, and combined with high-resolution image acquisition equipment, they can generate high-precision status data; tag status data is used as input to provide a high-quality data foundation for subsequent anomaly detection and circulation status prediction; smart material tags do not require external energy support, have low manufacturing costs, and are suitable for large-scale circulation traceability of agricultural products.

[0141] In the example, a cold chain logistics company transported a batch of refrigerated cherries. To ensure that the temperature and light conditions during transportation met the cold chain requirements, a smart material label was placed in each transport box. In a refrigerated environment, the color of the thermochromic microparticle film gradually changed from blue to lavender, reflecting the temperature rising from 2°C to 8°C; the high-resolution imaging device scanned every 5 minutes to generate label status data.

[0142] At the same time, the spectral response coating records the wavelength distribution of ambient light. Under direct illumination, the spectral data shows that the blue light component is enhanced, indicating that the light intensity is high. The generated status data is uploaded to the central data processing system, combined with the anomaly detection algorithm to identify the risk of excessive temperature during transportation and send an alarm.

[0143] The temperature and light data recorded by the smart material labels clearly reflect the environmental conditions of cold chain transportation, helping logistics companies to adjust refrigeration equipment in time before problems occur, thus avoiding damage to the quality of cherries; consumers can view complete cold chain condition information through the traceability platform, enhancing their trust in product safety.

[0144] Preferably, the step of performing anomaly detection on dynamically bound data includes:

[0145] Combined with the improved isolation forest algorithm, by constructing a random partition tree of multidimensional data, the temperature and humidity values, texture feature values ​​and time series recorded in the label status data are segmented layer by layer; the distribution density and the degree of outlier of data points in each segmented area are analyzed, anomalies are identified and anomaly detection results are generated, which include the occurrence time, impact range and outlier score of the anomaly.

[0146] The dynamically bound tag status data includes temperature and humidity values, texture feature values ​​and their time series information, which are used to describe the environmental status of agricultural products at different times during the circulation process. These data are collected in real time through smart material tags and dynamically updated to the digital twin.

[0147] Tag status data is represented in time series form:

[0148]

[0149] in, is the timestamp, is the temperature value, is the illumination feature value. By merging the data at multiple time points, a complete input matrix is ​​formed:

[0150]

[0151] This matrix contains all the feature dimensions in the time series.

[0152] The isolation forest algorithm is an unsupervised anomaly detection method based on a random partition tree. Its core idea is that the path length of outliers in the random partition tree is shorter because they are easily isolated. Based on the traditional isolation forest algorithm, the present invention introduces a dynamic weight adjustment mechanism to improve the detection accuracy in multidimensional time series data. Random partition tree construction: 1. Randomly select a data dimension (such as temperature, light or time); 2. Randomly select a split point in the selected dimension to divide the data into two parts; 3. Recursively perform the above operations until the data set is completely split or the preset maximum tree depth is reached; 4. Output the path length of each data point in the tree. The formula represents the path length of a single point as:

[0153]

[0154] in, is the data point, is the depth of the tree, Indicates that the data point is The length of the path in the subdivision.

[0155] Calculate the outlier score for each data point based on the normalized result of the path length:

[0156]

[0157] in, is the normalization coefficient, is the total number of data points. The outlier score ranges from 0 to 1, and the higher the score, the more likely it is an outlier.

[0158] Dynamically adjust the partition weights based on the importance of feature dimensions. For example, when temperature fluctuations are small, increase the partition weight of the light dimension to more sensitively detect light anomalies; and give higher priority to abnormal points with short-term mutations based on trend changes in the time dimension.

[0159] The algorithm analyzes the dynamic binding data and generates anomaly detection results, including the occurrence time, impact range and outlier score of the anomaly point:

[0160] Occurrence time: the specific timestamp of the abnormal point in the time series (such as ).

[0161] Impact range: defines the characteristic dimensions involved in the abnormal point (such as temperature, light) and the magnitude of the abnormality.

[0162] Outlier Scoring: Assign a score to each outlier point, where points with high scores are more likely to be serious outliers.

[0163] The improved isolation forest algorithm can quickly identify anomalies in the circulation process, including sudden changes in temperature and humidity and abnormal light, ensuring the storage and transportation conditions of agricultural products; through the dynamic weight adjustment mechanism, the algorithm performs more accurately in multidimensional time series data and can capture subtle anomalies in complex environmental changes; the anomaly detection results not only support real-time early warnings, but also provide a reliable data basis for subsequent circulation traceability and status prediction.

[0164] In an embodiment, when transporting mangoes, an agricultural product logistics company hopes to monitor the temperature, humidity and light conditions in the transport box to prevent the quality of the fruit from deteriorating due to abnormal environment.

[0165] Smart material tags are installed in the transport boxes to record temperature, humidity and light data in real time, and dynamically bind the data to the digital twin of Mango. The tag status data is uploaded to the central system via the wireless network to form a time series matrix: The data was analyzed using an improved isolation forest algorithm, which detected an abnormal increase in light intensity during a certain period of time, with an outlier score of 0.95, which was marked as an anomaly. At the same time, temperature records showed large fluctuations, deviating from the normal range.

[0166] The generated anomaly detection results show that the anomaly occurred in the fifth hour of transportation, and the abnormal value of light intensity affected three transport boxes.

[0167] The logistics company adjusted the light shielding measures of the transport boxes based on the abnormal detection results and optimized the operating parameters of the cold chain equipment to avoid the problem of mangoes being over-ripe due to excessive light. Consumers can check the environmental status during transportation through the traceability system and confirm the controllability of the quality of mangoes.

[0168] Preferably, the improved isolation forest algorithm adopts a dynamic adjustment strategy for the random partitioning of label status data, sets the partition depth and the number of partition trees according to the non-uniformity of data distribution, and determines the credibility of outliers by calculating the cumulative distribution probability of outliers.

[0169] In the circulation of agricultural products, tag status data (such as temperature, humidity, and light intensity) usually exhibits non-uniform distribution characteristics. For example, some transport boxes in cold chain logistics may experience greater temperature and humidity changes due to insufficient shielding, while the environment of other boxes may be relatively stable. This non-uniformity requires the detection algorithm to have flexible partitioning capabilities to accurately process data in high-volatility and low-volatility areas. Input data: dynamically bound tag status data, recorded as a multidimensional time series:

[0170]

[0171] in, is the timestamp, is the temperature value, is the illumination characteristic value.

[0172] The core of the Isolation Forest Algorithm is the random partition tree. In traditional methods, the depth and number of partition trees are fixed, which may lead to a decrease in detection accuracy in high-density areas when facing non-uniformly distributed data. The present invention optimizes the partition process by dynamically adjusting the strategy:

[0173] The partition depth is dynamically set according to the local density of data distribution. For areas with high distribution density, deeper partitions are used to refine the location of outliers; for areas with sparse distribution, shallower partitions are used to reduce computational complexity.

[0174] Formula Description: Partition Depth Based on the local density of data points Make adjustments:

[0175]

[0176] in, is the adjustment factor, is the local density, indicating the data point The number of nearby points.

[0177] Adjust the number of partition trees according to the degree of outliers in the overall data. Increase the number of partition trees when there are many outliers to improve the robustness of detection; reduce the number of partition trees when the data is evenly distributed to reduce the computational cost. Calculate the number of partition trees required :

[0178]

[0179] in, is the adjustment factor.

[0180] The credibility of outliers is an important output indicator of anomaly detection. This paper calculates the credibility of outliers through cumulative distribution probability and quantifies the reliability of detection results. The abnormal probability distribution of data points is calculated based on the outlier score:

[0181]

[0182] in, For data points The abnormal credibility of Give the outlier score for this point, is the probability density function of the score. The higher the value, the more likely the data point is an outlier. , marking high-confidence outliers.

[0183] By dynamically adjusting the partition depth and the number of partition trees, the present invention can flexibly adapt to the data distribution characteristics of different regions and improve the overall accuracy of anomaly detection; the credibility of anomaly points is calculated by cumulative distribution probability to provide a quantitative reliability evaluation for the detection results, which is convenient for result verification in the actual traceability system; the dynamic partitioning strategy optimizes resource allocation according to data distribution, significantly reducing the computing cost while ensuring detection accuracy.

[0184] Example: A cold chain transportation system is used to distribute a batch of cherries. During transportation, it is necessary to detect abnormal fluctuations in ambient temperature and light intensity to ensure cold chain conditions. The temperature and light intensity are recorded through smart material tags, and multi-dimensional time series data is generated: The data is uploaded to the central processing system.

[0185] The data was analyzed using an improved isolation forest algorithm: the partition depth was adjusted dynamically, and deeper partitions were used in areas of the transport box with large temperature fluctuations for refined detection; the number of partition trees was adjusted according to the outlier rate of the global data, and the number of trees was increased to enhance robustness when the outlier rate was high during transportation; the credibility of outliers was calculated through the cumulative distribution probability, and the abnormal time points and the scope of abnormal impact were marked.

[0186] Abnormal result: The system detected a rapid rise in temperature in a transport box in the fourth hour (outlier score of 0.98, credibility of 95%), and issued an early warning. The improved isolation forest algorithm accurately identified temperature anomalies during transportation, providing support for logistics personnel to adjust cold chain equipment in a timely manner, effectively avoiding the problem of cherries rotting due to excessive temperature. At the same time, consumers can view the environmental status during transportation through the traceability system, which enhances their trust in cold chain safety.

[0187] Based on the digital twin and anomaly detection results, data association analysis is performed, and the reconstructed digital twin is generated by optimizing the structure of the digital twin. A recursive neural network model combined with a multivariate interpolation algorithm is used to complete the anomaly points and predict the future state to generate predicted state data.

[0188] Digital twins are digital representations of agricultural products in virtual space, storing the physical characteristics, environmental status, and circulation path information of agricultural products. By analyzing the features of abnormal points in the anomaly detection results (such as time, dimension, and outlier score), the structure of the twin can be optimized to generate a reconstructed digital twin containing key abnormal nodes and associated paths.

[0189] According to the results of the association analysis, add state dimensions related to the abnormal points (such as abnormal time range and impact path). Remove redundant data to ensure that the twin structure is concise and can effectively reflect the circulation status. Generate a reconstructed digital twin.

[0190] The multivariate interpolation algorithm is used to fill in the missing data near the outlier point, and the recursive neural network model is used to learn the dynamic characteristics of the time series data and predict the future state change trend. The multivariate interpolation algorithm is used to complete the data around the outlier point, such as using the Lagrange interpolation method to smooth the time series data. The completed data is saved in the reconstructed twin.

[0191] The recurrent neural network (RNN) model is used to learn time series data and capture the dynamic laws of changes in temperature, humidity, light, etc. RNN can effectively extract the contextual relationship of time series data through hidden layer recursive connections. After model training, the state prediction results for future time periods are generated, including temperature and humidity trends and light changes.

[0192] The predicted status data includes the environmental status change trend and potential abnormal points in the future, which is used to assist in risk warning in the circulation link. Combined with the prediction results and abnormal threshold settings, possible abnormal points in the future are marked.

[0193] Preferably, the step of reconstructing the digital twin includes:

[0194] By adding associated dimensions to the structure of the digital twin, including the spectral change characteristics of the environmental state and the temperature and humidity change trends; combining the multivariate interpolation algorithm to complete the data near the abnormal points, and using the Lagrange interpolation method to smooth the values ​​of the continuous time series, the completed time series data is generated, and the digital twin is reconstructed based on this.

[0195] Digital twins are digital mappings of agricultural products in virtual space, including their physical characteristics, environmental status, and spatial location. In the circulation process, in order to more accurately reflect dynamic changes, it is necessary to add associated dimensions based on the results of anomaly detection to make the twin structure more adaptable. The newly added dimensions include: spectral change characteristics of environmental status, which analyzes environmental status changes based on spectral data, such as fluctuations in light intensity and spectral reflectance values, and provides a refined description of environmental changes; temperature and humidity change trends, which capture abnormal trends in the cold chain environment by analyzing the changing trends of temperature and humidity values ​​in time series.

[0196] The environmental state change characteristics are extracted from the spectral data. The collected spectral data is used to calculate the spectral offset and peak change to generate characteristic values ​​reflecting the environmental state change:

[0197]

[0198] in, represents the shift of the spectral peak, is the current spectrum peak, is the reference spectrum peak.

[0199] Calculate the temperature and humidity change rate and add the twin. The temperature and humidity change rate calculation formula is:

[0200]

[0201]

[0202] in, and are the temperature and humidity change rates, , For the The temperature and humidity values ​​at a time point, is the time interval.

[0203] In the process of anomaly detection, missing data or discontinuous data may occur. The data near the anomaly point is supplemented by a multivariate interpolation algorithm, and the time series is smoothed by Lagrange interpolation to ensure the continuity and accuracy of the data.

[0204] The interpolation algorithm is used to fill in the abnormal point data in the time dimension. The input is the valid data before and after the abnormal point, and the output is the completed intermediate data.

[0205] For continuous time series data, Lagrange interpolation method is used for smoothing, the formula is:

[0206]

[0207] in, is the interpolation function, is the known data point value, and are the corresponding time point coordinates respectively. A smooth data stream is generated by interpolation.

[0208] The optimized twin structure includes newly added associated dimensions and completed time series data, which enables it to fully record the dynamic status of agricultural products. The reconstructed digital twin is represented as:

[0209]

[0210] in, For the texture characteristics of agricultural products, is the environmental state, For geographical location, For the newly added spectral change characteristics and temperature and humidity change trend dimensions, For time series.

[0211] The newly added associated dimension enables the digital twin to more comprehensively reflect the dynamic environmental changes of agricultural products and enhances the data's expressiveness. Through interpolation completion and time series smoothing, the problem of missing and discontinuous data in anomaly detection is solved, providing reliable data input for subsequent traceability analysis; the reconstructed twin provides a complete state description, laying the foundation for anomaly location and future state prediction in circulation traceability.

[0212] In the implementation example, when a cold chain logistics company is transporting bananas, it needs to monitor the dynamic changes of the cold chain environment by reconstructing the digital twin and complete the missing data caused by sensor failure.

[0213] Extract environmental state change characteristics (such as spectral offset of light changes) from spectral data. Calculate temperature and humidity change trends and add them as new dimensions to the twin.

[0214] The multivariate interpolation algorithm was used to complete the missing temperature and humidity data between the 4th and 5th hours. The Lagrange interpolation method was used to smooth the time series from the 3rd to the 6th hour to ensure data continuity.

[0215] Integrate the completed and optimized time series data into the twin to generate a reconstructed twin.

[0216] The reconstructed digital twin comprehensively records the dynamic changes of environmental conditions during transportation, including spectral change characteristics and temperature and humidity change trends; it completes missing data, solves the problem of data discontinuity, and provides logistics companies with high-precision environmental monitoring and anomaly analysis tools.

[0217] Preferably, Figure 3 As shown, the steps of future state prediction include:

[0218] The recursive neural network model is used to learn the completed time series data and build a prediction model. By inputting a continuous environmental state time series, the predicted state data for the future time period is generated. The predicted state data includes the temperature and humidity trends of the future environment, changes in spectral characteristics, and the location and impact range of potential abnormal points.

[0219] The completed time series data is used as the input of the prediction model, including the changes in environmental conditions during the circulation of agricultural products. Through multivariate interpolation algorithms and outlier processing, the continuity and integrity of the time series are ensured, providing high-quality data for the training of the recursive neural network model. Data format:

[0220]

[0221] in, is the timestamp, is the temperature value, is the texture feature value.

[0222] Recurrent neural network (RNN) is a neural network structure suitable for time series modeling. It can capture the contextual relationship of time series data through hidden states. The RNN model predicts the changing trend of future states by learning the temporal dynamics of environmental state data. Model structure:

[0223] Input layer: accepts multidimensional time series data, and the input is the environmental state value .

[0224] Hidden layer: The recursively connected hidden layer captures the dynamic characteristics of the time series. The update formula of the hidden state is

[0225]

[0226] in, is the current hidden state, is the hidden state of the previous time step, is the current input data, , is the weight matrix, is the bias, is the activation function (such as ReLU or Tanh).

[0227] Output layer: predict future temperature, humidity and texture feature values.

[0228] Loss function: Use mean square error (MSE) as the objective function to optimize the prediction accuracy of the model:

[0229]

[0230] in, is the true value, is the predicted value, is the number of samples.

[0231] Based on the trained RNN model, the environment state changes in the future are predicted by inputting the completed time series data. Prediction output:

[0232] Temperature and humidity trends: Temperature and humidity values ​​at future time points, reflecting the changing trend of environmental conditions.

[0233] Spectral feature changes: Texture feature values ​​for future time periods, capturing the impact of light changes on agricultural products.

[0234] Potential anomalies: By comparing with the anomaly threshold, possible anomalies in the future and their impact range are marked.

[0235] The results show that:

[0236]

[0237] in, For future time points, , are the predicted temperature, humidity and spectrum values.

[0238] The recursive neural network is used to capture the temporal dynamics of environmental conditions and generate high-precision predictions of future temperature and humidity trends and spectral changes. The potential anomalies marked in the prediction results and their impact range provide logistics companies with a basis for making decisions on adjusting cold chain equipment and optimizing transportation conditions in advance. The completed time series is combined with the prediction model to fully tap the value of historical data and provide continuous protection for dynamic monitoring and decision-making support in the circulation process.

[0239] Example: A cold chain transportation company for agricultural products needs to make future predictions on the environmental status during cherry transportation to identify potential risks and optimize transportation conditions. Implementation process:

[0240] The temperature, humidity and spectrum data are recorded by smart material tags, and the missing values ​​are supplemented by interpolation algorithms to generate continuous time series data:

[0241] The RNN model is trained using historical transportation data to capture the temporal dynamic characteristics of the environmental state. During the training process, the mean square error is used to optimize the model to ensure the prediction accuracy of temperature, humidity and spectral characteristics.

[0242] Input the completed time series data, predict the temperature and humidity trends and spectral feature changes in the next 12 hours, and generate the prediction results:

[0243] By comparing with the abnormal threshold, it was found that the temperature in the 8th hour might rise to 15°C (outside the cold chain temperature range) and was marked as a potential abnormal point.

[0244] According to the prediction results, the logistics company adjusted the operating parameters of the cold chain equipment in the 6th hour and successfully resolved the abnormal temperature risk in the 8th hour, ensuring the quality of the cherries; consumers can view the temperature and humidity trends and spectral changes during the transportation process through the traceability platform, which enhances their trust in product quality.

[0245] like Figure 4 As shown, the reconstructed digital twin data and predicted status data are extracted to construct a knowledge graph including the status of agricultural products, circulation paths and future predicted status; based on the knowledge graph, a traceability path including historical status, current status and future risk warnings is generated.

[0246] Preferably, the construction of the knowledge graph includes:

[0247] Nodes and associated paths containing agricultural product status information are extracted from the reconstructed digital twin data, and nodes are defined as status types or location information, and paths are defined as associations between data. Combined with the predicted status data, the future status is mapped into virtual nodes in the knowledge graph, and a knowledge graph containing historical status, current status, and future status association paths is generated, which is converted into structured data based on the resource description framework for storage and query.

[0248] The reconstructed digital twin data contains the dynamic status information of agricultural products, including environmental parameters (such as temperature, humidity, spectral characteristics) and spatial information (such as geographic location). The predicted status data provides the trend of future status changes and potential abnormal points. By extracting and classifying these data, the node types in the knowledge graph can be defined. Node definition:

[0249] The state node represents the historical state, current state, and future state of agricultural products; the historical state and current state come from the temperature and humidity data, spectral change characteristics, and geographic location in the twin.

[0250] The future state is the predicted state data, including temperature and humidity trends and potential abnormal points.

[0251] Location nodes, which describe the geographical location of agricultural products during circulation, are defined based on the geographic tag information of the twin.

[0252] Example of a state node extracted from a digital twin:

[0253]

[0254] in, They are temperature, texture feature value and timestamp respectively; example of location node extracted from geographic information:

[0255]

[0256] in, , is the longitude and latitude.

[0257] The association path is used to describe the relationship between nodes, including the time association between state nodes, the spatial association between position nodes, and the cross-dimensional association between state and position. Path definition:

[0258] Time path: represents the continuity of agricultural product status changes over time.

[0259] Spatial path: represents the geographical movement trajectory of agricultural products during the circulation process.

[0260] State-location path: represents the relationship between the state and the geographic location at a specific point in time.

[0261] Time Path:

[0262] Space Path:

[0263] Status-Location Path:

[0264] The predicted state data is used to generate virtual nodes to map environmental changes and potential risks in the future state. By associating virtual nodes with historical and current nodes, a complete knowledge graph containing future dimensions is formed. Virtual node mapping, the virtual nodes generated by the predicted state data include temperature and humidity trends, spectral feature changes, and abnormal point information at future time points. For example:

[0265]

[0266] The knowledge graph is represented by a collection of nodes and paths:

[0267]

[0268] in, Including historical, current and future state nodes, Includes time path, space path and state-position path.

[0269] Converted into structured data, the knowledge graph is represented as triples using the Resource Description Framework (RDF) for easy storage and query:

[0270] Example triple:

[0271] <Agricultural product status 1, time association, agricultural product status 2>

[0272] <Agricultural product location 1, spatial association, agricultural product location 2>

[0273] <Agricultural product status 1, associated location, agricultural product location 1>

[0274] Through the nodes and paths of the knowledge graph, the historical status, current status and future status are integrated to achieve a dynamic description of the entire life cycle of agricultural products; based on the knowledge graph, traceability paths can be efficiently generated, containing comprehensive information on time, space and environmental status, providing regulators and consumers with traceable and transparent data; the virtual nodes generated by future state mapping provide prediction results of environmental changes and potential risk warnings, providing technical support for early warning and optimized decision-making of logistics companies.

[0275] Example: An organic fruit and vegetable supply chain company needs to build a traceability system to monitor the circulation status of agricultural products in real time and generate traceability paths and risk warnings through knowledge graphs. Implementation process:

[0276] Extract the reconstructed digital twin data and generate nodes containing historical and current states. Generate virtual nodes for future states from predicted state data.

[0277] Define the time path, space path and state-location path, and associate the relationship between nodes. For example, the temperature and humidity history records of a fruit or vegetable are associated with the geographic location information through the time path and space path to form a circulation trajectory.

[0278] The node and path collections are converted into a knowledge graph structure and stored in a graph database using the RDF format.

[0279] Based on the knowledge graph, the complete circulation process of certain fruits and vegetables can be queried, including the historical environmental status, current circulation location and future temperature and humidity trends, and a traceability path with risk warnings can be generated.

[0280] Enterprises can clearly understand the circulation process and environmental changes of fruits and vegetables through knowledge graphs; consumers can check the source of products, transportation conditions and future risk warnings, which enhances their trust in product quality; logistics companies optimize cold chain operation parameters and reduce product losses through risk warnings based on future status predictions.

[0281] like Figure 5 As shown, a system for tracing the origin of agricultural products is used to implement the method for tracing the origin of agricultural products. The system includes:

[0282] A data acquisition module, used to obtain RGB image data, environmental data and geographic location information of agricultural products, the data acquisition module includes an RGB imaging device, a temperature and humidity sensor, a gas sensor and a high-precision geographic positioning device to collect texture feature data, temperature and humidity data, gas concentration data and geographic tag data of agricultural products;

[0283] A digital twin generation module is used to integrate the RGB image data, environmental data and geographic tag data obtained by the data acquisition module to generate a digital twin containing a unique identifier, environmental status and geographic location;

[0284] A state recording module, including a smart material tag and a state acquisition device. The smart material tag is used to record temperature, humidity, light changes and other environmental conditions in real time. The state acquisition device is used to extract state data from the smart material tag and dynamically bind it to the digital twin.

[0285] An anomaly detection module is used to detect anomalies in the environment state based on the dynamic binding data generated by the state recording module in combination with the improved isolation forest algorithm, and generate anomaly detection results, wherein the anomaly detection results include the occurrence time, dimension and outlier score of the anomaly;

[0286] The twin reconstruction and prediction module is used to optimize the structure of the digital twin according to the anomaly detection results to generate the reconstructed digital twin, and to complete the data near the anomaly point by combining the multivariate interpolation algorithm, and to analyze the time series data through the recursive neural network model to generate the predicted status data;

[0287] The knowledge graph construction module is used to extract key nodes and associated paths from the reconstructed digital twin data and predicted state data, construct a knowledge graph containing historical states, current states, and future predicted states, and generate traceability path data;

[0288] The feedback generation module is used to generate customized feedback reports for consumers, producers and regulatory agencies based on the knowledge graph. The feedback reports include the traceability path of agricultural products, status history records and future risk warnings.

[0289] This system focuses on the core goal of agricultural product circulation traceability, integrating multiple technologies such as multimodal data collection, digital twin generation and dynamic monitoring, anomaly detection and state prediction, and knowledge graph construction to form a full-process intelligent traceability system. Its main modules and principles are as follows:

[0290] Data acquisition module: The texture features of the surface of agricultural products are acquired through RGB imaging equipment, the environmental status is collected by combining temperature and humidity sensors and gas sensors, and high-precision geo-positioning equipment is used to record geo-tagged data to ensure the integrity and multi-dimensionality of the collected data.

[0291] Digital twin generation module: Integrate the collected data to generate a digital twin, which serves as the unique digital mapping of agricultural products in the virtual space, including the environmental status, texture characteristics and geographic location of the agricultural products.

[0292] Status recording module: Smart material tags record dynamic environmental conditions such as temperature, humidity, and light in real time, and dynamically bind the data to the digital twin through status acquisition devices to form a status record that is updated over time.

[0293] Anomaly detection module: Combined with the improved isolation forest algorithm, it analyzes abnormal points in dynamic binding data, marks the abnormal time, dimension and outlier score, and provides accurate support for anomaly identification in the circulation process.

[0294] Twin reconstruction and prediction module: Optimize the twin structure based on the anomaly detection results, complete the missing data through multivariate interpolation, use the recursive neural network model to predict the future environmental state, and generate prediction data including temperature and humidity trends and potential anomalies.

[0295] Knowledge graph construction module: Extract key nodes and paths from the reconstructed digital twin and prediction data to generate a knowledge graph containing historical, current and future states for storage and query.

[0296] Feedback generation module: Generates personalized feedback reports for consumers, producers and regulators based on the knowledge graph, including traceability paths, status records and future risk warnings, to achieve transparent traceability and intelligent early warning.

[0297] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0298] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for tracing the origin of agricultural products, characterized in that: The following steps are involved: Obtaining RGB image data, environmental data, and geographic location information of the agricultural product, and integrating all the data to generate a digital twin, wherein the digital twin includes a unique identifier of the agricultural product, a recorded environmental status, and a geographic location; In the circulation of agricultural products, smart material tags are used to record changes in temperature, humidity and light in real time, and the status data recorded by the tags are dynamically bound to the digital twin. Anomaly detection is performed on the dynamically bound data, and the improved isolation forest algorithm is used to identify abnormal points in the environmental state to generate anomaly detection results. Based on the digital twin and anomaly detection results, data association analysis is performed, and the reconstructed digital twin is generated by optimizing the structure of the digital twin. A recursive neural network model combined with a multivariate interpolation algorithm is used to complete the anomaly points and predict the future state to generate predicted state data. The step of generating a reconstructed digital twin by optimizing the structure of the digital twin includes: adding associated dimensions to the structure of the digital twin, including the spectral change characteristics of the environmental state and the temperature and humidity change trends; completing the data near the abnormal point in combination with a multivariate interpolation algorithm, smoothing the values ​​of the continuous time series using a Lagrange interpolation method, generating completed time series data, and reconstructing the digital twin based on the data; The steps of future state prediction include: The recursive neural network model is used to learn the completed time series data and build a prediction model; by inputting a continuous environmental state time series, the predicted state data for the future time period is generated, and the predicted state data includes the temperature and humidity trends of the future environment, the changes in spectral characteristics, and the location and impact range of potential abnormal points; Extract the reconstructed digital twin data and predicted status data, and build a knowledge graph that includes the status of agricultural products, circulation paths, and future predicted status; generate a traceability path based on the knowledge graph that includes historical status, current status, and future risk warnings.

2. The agricultural product circulation traceability method according to claim 1, characterized in that: The steps of generating a digital twin include: The RGB image data of agricultural products is obtained through RGB imaging equipment, and the edge features in the RGB image data are processed using the morphological gradient enhancement algorithm to highlight the texture features of leaf veins or fruit surfaces. The processed texture feature data is embedded in combination with the deep contrast learning model to generate the biological texture identification data of each agricultural product, and the biological texture identification data is integrated with the environmental data and geographic location information to generate a digital twin.

3. The agricultural product circulation traceability method according to claim 1, characterized in that: The environmental data includes temperature and humidity data and gas concentration data. The temperature and humidity data are collected by temperature and humidity sensors, and the gas concentration data is obtained by detecting the concentration level of ethylene gas. The geographic location information is obtained by high-precision GNSS equipment, and the data in weak signal areas is corrected based on a dynamic geographic tagging algorithm.

4. The agricultural product circulation traceability method according to claim 1, characterized in that: The smart material label includes a thermochromic particle film and a spectrally responsive coating, wherein the thermochromic particle film changes color within a specific threshold range according to changes in ambient temperature, and the spectrally responsive coating reflects light signals of specific wavelengths under different lighting conditions; the color changes and spectral reflectance characteristics of the label are recorded by a high-resolution image acquisition device, and real-time label status data is generated.

5. The agricultural product circulation traceability method according to claim 1, characterized in that: The step of performing anomaly detection on dynamically bound data includes: Combined with the improved isolation forest algorithm, by constructing a random partition tree of multidimensional data, the temperature and humidity values, texture feature values ​​and time series recorded in the label status data are segmented layer by layer; the distribution density and the degree of outlier of data points in each segmented area are analyzed, anomalies are identified and anomaly detection results are generated, which include the occurrence time, impact range and outlier score of the anomaly.

6. The agricultural product circulation traceability method according to claim 5, characterized in that: The improved isolation forest algorithm adopts a dynamic adjustment strategy for the random partitioning of label state data, sets the partition depth and the number of partition trees according to the non-uniformity of data distribution, and determines the credibility of outliers by calculating the cumulative distribution probability of outliers.

7. The agricultural product circulation traceability method according to claim 1, characterized in that: The construction of the knowledge graph includes: Nodes and associated paths containing agricultural product status information are extracted from the reconstructed digital twin data, and nodes are defined as status types or location information, and paths are defined as associations between data. Combined with the predicted status data, the future status is mapped into virtual nodes in the knowledge graph, and a knowledge graph containing historical status, current status, and future status association paths is generated, which is converted into structured data based on the resource description framework for storage and query.

8. An agricultural product circulation traceability system, used to implement the agricultural product circulation traceability method according to any one of claims 1 to 7, characterized in that: The system includes: A data acquisition module, used to obtain RGB image data, environmental data and geographic location information of agricultural products, the data acquisition module includes an RGB imaging device, a temperature and humidity sensor, a gas sensor and a high-precision geographic positioning device to collect texture feature data, temperature and humidity data, gas concentration data and geographic tag data of agricultural products; A digital twin generation module is used to integrate the RGB image data, environmental data and geographic tag data obtained by the data acquisition module to generate a digital twin containing a unique identifier, environmental status and geographic location; A state recording module, including a smart material tag and a state acquisition device. The smart material tag is used to record temperature, humidity and light changes in real time. The state acquisition device is used to extract state data from the smart material tag and dynamically bind it to the digital twin. An anomaly detection module is used to detect anomalies in the environment state based on the dynamic binding data generated by the state recording module in combination with the improved isolation forest algorithm, and generate anomaly detection results, wherein the anomaly detection results include the occurrence time, dimension and outlier score of the anomaly; The twin reconstruction and prediction module is used to optimize the structure of the digital twin according to the anomaly detection results to generate the reconstructed digital twin, and to complete the data near the anomaly point by combining the multivariate interpolation algorithm, and to analyze the time series data through the recursive neural network model to generate the predicted status data; The knowledge graph construction module is used to extract key nodes and associated paths from the reconstructed digital twin data and predicted state data, construct a knowledge graph containing historical states, current states, and future predicted states, and generate traceability path data; The feedback generation module is used to generate customized feedback reports for consumers, producers and regulatory agencies based on the knowledge graph. The feedback reports include the traceability path of agricultural products, status history records and future risk warnings.

Images