Method for generating agricultural production process data

Abstract

The invention relates to the field of agricultural credit approval, and discloses an agricultural production process data generation method, which comprises the steps of receiving credit application data of a target farmer through a farmer terminal; querying a user file according to the peasant household attribute information to obtain a peasant household integrity score, a planting type, an Internet of Things crop monitoring equipment identifier and a key evidence chain storage file; if the integrity score does not meet the preset condition, verifying the authenticity of the asset ownership proof file through the key evidence chain storage file; after the authenticity verification is passed, acquiring corresponding agricultural production process data according to the planting type, forming agricultural production process display data, pushing the agricultural production process display data to an intelligent terminal for visual presentation, initiating interactive inquiry, and acquiring supplementary description information of the farmer; and receiving the supplementary description information and carrying out associative storage on the supplementary description information and agricultural production process data to generate an agricultural production process data evidence material. Therefore, the authenticity, the integrity and the verifiability of the data are improved.

Description

Technical Field

[0001] This invention relates to the field of agricultural informatization and intelligent management technology, specifically to a method for generating agricultural production process data. Background Technology

[0002] With the rapid development of information technology, agricultural production is gradually moving towards data-driven and intelligent processes. Financial institutions are increasingly demanding credit, insurance, and other services from the agricultural sector. However, existing technologies face significant technical challenges in acquiring, verifying, and utilizing farmers' authentic agricultural production information. The following technical problems frequently arise in the existing verification process of agricultural data: First, existing technologies lack a high-precision, closed-loop verification data processing architecture, making it difficult to efficiently integrate and structure multi-source agricultural production data; the lack of intelligent anomaly identification and guided interaction mechanisms makes it difficult to effectively correlate and corroborate objective monitoring data with farmers' subjective supplementary information, resulting in insufficient reliability of agricultural production process data. Second, existing technologies lack high-precision remote sensing data calibration and multi-source agricultural data processing methods for farmers' plots, making it difficult to accurately extract plot boundaries, identify abnormal conditions, and quantify disease characteristics for yield prediction. Summary of the Invention

[0003] The summary section of this invention provides a brief overview of the concepts, which will be described in detail in the detailed description section below. This summary section is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0004] This invention proposes a method for generating agricultural production process data to solve one or more of the technical problems mentioned in the background section above.

[0005] This invention provides a method for generating agricultural production process data, characterized by comprising: The system receives credit application data from target farmers through farmer terminals. The credit application data includes the target farmers' attribute information, asset ownership documents, and the reasons and amount of the credit application. Based on the attribute information, query the pre-built user profile to obtain the target farmer's credit score, planting type, the identification of the IoT crop monitoring equipment pre-configured by the farmer, and the evidence documents stored on the key evidence chain; If a farmer's credit score does not meet the preset conditions, the authenticity of the asset ownership certificate will be verified through the evidence documents stored on the key evidence chain. If the authenticity verification passes, the corresponding agricultural production process data is obtained according to the planting type; the agricultural production process data is then structured according to a preset data visualization format to form the agricultural production process display data for the target farmer. The agricultural production process data is pushed to the smart terminal in the business hall so that the smart terminal can visualize the data and initiate interactive inquiries with the target farmers to obtain supplementary information from the target farmers on key production nodes, abnormal growth stages or special agricultural activities. Receive supplementary information and link it with agricultural production process data for storage, generating supporting materials for agricultural production process data.

[0006] Optionally, based on the planting type, obtain the corresponding agricultural production process data, including: If the planting type represents open-field planting, then based on the land location information extracted from the asset ownership certificate and the attribute information of the target farmer, the standard remote sensing data marked with ownership is queried to obtain the marked plot area and plot area information corresponding to the land location information in the standard remote sensing data; and based on the land location information, real-time remote sensing datasets of multiple time points within the area are pulled, and based on the real-time remote sensing datasets, marked plot areas and plot area information, agricultural production process data are determined; If the planting type represents greenhouse planting, then query the crop growth environment data corresponding to the IoT crop monitoring device identifier within the target historical time period, and generate agricultural production process data.

[0007] Optionally, each real-time remote sensing data point in the real-time remote sensing dataset corresponds to a specific time point; Based on real-time remote sensing datasets, calibrated plot areas, and plot area information, agricultural production process data are determined, including: Using an instance segmentation model, image segmentation is performed on each real-time remote sensing data in the real-time remote sensing dataset to obtain multiple candidate polygons included in each real-time remote sensing data. The boundary of the marked area is defined as a standard polygon. The similarity between the standard polygon and multiple candidate polygons is calculated, and the top N candidate polygons with the highest similarity ranking for each real-time remote sensing data are selected. The candidate polygon with the highest frequency among N candidate polygons corresponding to multiple time points is selected as the target polygon. The region corresponding to the target polygon in each real-time remote sensing data is cropped to obtain multiple remote sensing image regions corresponding to different time points as agricultural production process data.

[0008] Optionally, the IoT crop monitoring equipment identified by the IoT crop monitoring equipment identifier includes soil environment monitoring sensors, crop growth monitoring sensors, and greenhouse cameras; and If the planting type represents greenhouse cultivation, then query the crop growth environment data corresponding to the IoT crop monitoring device identifier within the target historical time period, and generate agricultural production process data, including: If the planting type represents greenhouse planting, then query the data from soil environmental monitoring sensors, crop growth monitoring sensors, and greenhouse cameras within the target historical time period and identify them as agricultural production process data.

[0009] Optionally, the method for generating agricultural production process data according to the present invention further includes: If the planting type represents greenhouse planting, then the greenhouse images collected by the greenhouse camera are extracted according to the preset image extraction ratio to obtain the extracted greenhouse image set; according to the preset M time intervals and the collection timestamp of each greenhouse image, the extracted greenhouse image set is divided into M greenhouse image groups, and the M greenhouse image groups correspond one-to-one with the M time intervals. The extracted greenhouse image set is identified to obtain the greenhouse crop category. Based on the greenhouse crop category, a pre-configured disease feature database is queried to obtain at least one typical disease feature corresponding to the greenhouse crop category and the disease stage corresponding to each typical disease feature. Based on the disease stage corresponding to each typical disease characteristic, each typical disease characteristic is assigned to one or more time intervals.

[0010] Optionally, the method for generating agricultural production process data according to the present invention further includes: For each greenhouse image group, image features of each greenhouse image are extracted, and the average image feature within the group is calculated. Based on the Euclidean distance between the average image feature and each image feature, each greenhouse image group is divided into an abnormal greenhouse image group and a normal greenhouse image group. The average abnormal image feature corresponding to the abnormal greenhouse image group is determined, and its similarity is calculated with each corresponding typical disease feature. If the similarity is greater than or equal to a preset similarity threshold, the corresponding typical disease feature is identified as a potential disease feature. Based on the characteristics of potential diseases and the data sequence of crop growth indicators measured by crop growth monitoring sensors, a yield prediction value is generated. Add the yield forecast and the potential disease characteristics corresponding to each time interval to the agricultural production process data.

[0011] Optionally, based on the land location information extracted from the asset ownership certificate and the attribute information of the target farmer, the standard remote sensing data marked with ownership is queried to obtain the marked plot area and plot area information corresponding to the land location information in the standard remote sensing data, including: Based on the attribute information of the target farmers, the standard remote sensing data is queried to obtain the candidate calibration plot area and the corresponding candidate plot area information that match the user identifier of the target farmers; The candidate plot area information is checked for consistency with the land location information. If the check passes, the candidate plot area is determined as the designated plot area, and the candidate plot area information is determined as the plot area information.

[0012] Optionally, agricultural production process data can be pushed to smart terminals in the business hall, enabling the smart terminals to visualize the data and initiate interactive inquiries with target farmers. This allows the target farmers to provide supplementary information regarding key production nodes, abnormal growth stages, or special agricultural activities, including: Based on the data displayed in the agricultural production process, identify suspected abnormal features; annotate the suspected abnormal features to obtain annotation results; and generate corresponding query items based on the annotation results. The query items are sent to the smart terminal, which then presents the questions one by one on the interactive interface according to the priority of the query items, guiding the target farmers to provide supplementary explanations for the corresponding key production nodes, abnormal growth stages or special agricultural activities, and obtain supplementary explanation information.

[0013] The present invention has the following beneficial effects: 1. Data authenticity and reliability are ensured. Specifically, farmer terminals receive credit application data, including attribute information, asset certificates, application reasons, and loan amounts. Based on attribute information, user profiles are queried to obtain credit scores, planting types, IoT device identifiers, and key evidence documents. Asset certificates of farmers with low credit scores are verified for authenticity via blockchain to ensure data credibility. Corresponding agricultural production process data, including remote sensing images or greenhouse IoT monitoring data, is obtained based on planting type, achieving multi-source data integration. Farmer information, production process data, and supplementary explanations are linked and stored to form a complete data chain, which is then structured and visualized. The backend server analyzes and displays the data, identifying key production nodes, abnormal growth stages, and special agricultural activities. Supplementary explanations from farmers are collected through smart terminals to generate supporting materials for agricultural production process data, ensuring data authenticity, completeness, and verifiability, providing a reliable basis for credit approval and agricultural management.

[0014] 2. This technology enables accurate monitoring of crop growth processes, early disease risk quantification and warning, and precise yield prediction. Specifically, it extracts land location information and farmer attributes from asset ownership documents, queries candidate plots in a standard remote sensing database, and performs consistency verification through name or number comparison and spatial overlay calculation to achieve high-precision and verifiable plot labeling, ensuring accurate and reliable boundaries at different time points. For open-field cultivation, it identifies candidate polygons in real-time remote sensing data using an instance segmentation model, filters them based on similarity with labeled plot polygons, and generates agricultural production process data, achieving automated and time-series data acquisition. For greenhouse cultivation, this application divides historical data collected by soil environment, crop growth sensors, and cameras into time intervals and extracts and organizes images to form a structured dataset, providing a foundation for subsequent analysis. By extracting greenhouse image features and matching them with typical disease features, it achieves automatic identification and quantification of abnormal images and potential disease features, improving the accuracy of disease monitoring. Ultimately, by combining potential disease characteristics with time series of crop growth indicators, a yield prediction model is generated and the predicted values ​​are integrated into agricultural production data to achieve more accurate and dynamic yield prediction, providing reliable data support for agricultural management and credit risk control. Attached Figure Description

[0015] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0016] Figure 1 This is a flowchart of a method for generating agricultural production process data according to the present invention; Figure 2 This is the image segmentation result in a method for generating agricultural production process data according to the present invention; Figure 3 This is a schematic diagram of the intelligent terminal interface for a method of generating agricultural production process data according to the present invention. Detailed Implementation

[0017] The invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the drawings and embodiments of the invention are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0018] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0019] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0020] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0021] The names of messages or information exchanged between the various devices of this invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0022] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] like Figure 1 The diagram shows a flowchart of a method for generating agricultural production process data according to the present invention, which specifically includes the following steps: Step 101: Receive the credit application data of the target farmers through the farmer terminal. The credit application data includes the target farmers' attribute information, asset ownership documents, and the reasons and amount of the credit application. In some embodiments, the execution entity of the method for generating agricultural production process data according to the present invention is a backend server. The farmer terminal refers to a smart device operated by the farmer and equipped with network communication capabilities, such as a smartphone, tablet, or personal computer, running a dedicated credit application client (such as an app or webpage). Based on this, the backend server receives data (i.e., credit application data) transmitted from the farmer terminal through its open application programming interface (API) endpoint or web form submission interface. The target farmer specifically refers to the individual farmer or business entity submitting the credit application. Credit application data refers to all necessary information and documents submitted by the farmer for the credit approval process, forming the basis for subsequent approval. Among these, attribute information is data fields used to identify and describe the farmer's identity and basic situation, typically including: farmer's name, ID number, mobile phone number, home address, and contact information. Asset ownership certificates are electronic documents proving that the farmer has legal rights to the agricultural assets mortgaged or relied upon for the credit application, such as: land contract management right certificates, real estate ownership certificates (for fixed facilities such as greenhouses), and scanned copies or photos of agricultural machinery purchase invoices. The reason for the loan application is a statement filled out by the farmer explaining the specific purpose for which the loan funds are planned, such as "purchasing high-quality rice seeds and fertilizers" or "installing a drip irrigation system for 30 mu of orchards." The loan amount refers to the specific monetary amount the farmer wishes to borrow, such as "50,000 yuan."

[0024] Step 102: Based on the attribute information, query the pre-built user profile to obtain the target farmer's integrity score, planting type, the identifier of the IoT crop monitoring device pre-configured by the farmer, and the evidence files stored on the key evidence chain. In some embodiments, the pre-built user profile refers to an electronic profile database established and maintained by the backend server for registered farmers before receiving the application. This database is continuously updated and aggregates farmers' historical behavior and asset data. The backend server extracts key attribute information that uniquely identifies the farmer from the credit application data packet it receives, most typically the ID card number. Subsequently, the backend server uses the ID card number as the primary key or index key for a database query, executing a precise query statement in its internally maintained "User Profile" database table to obtain all pre-defined profile fields bound to that ID card number, including: farmer credit score, planting type, IoT device identifier, and on-chain evidence documents. The farmer credit score is a quantified value (e.g., 0-100 points) used to comprehensively assess the farmer's creditworthiness. This score may be calculated based on factors such as the farmer's historical repayment records, cooperative evaluations, and production and operational stability. The planting type categorizes the farmer's main production mode, such as "open-field planting" and "greenhouse planting." The IoT crop monitoring device identifier is a code used to uniquely identify IoT monitoring devices (such as sensors and cameras) within the farmer's plot. The key evidence chain on-chain evidence document is an evidence certificate obtained by storing the hash value of key information such as asset ownership certificates into the blockchain. This document proves the existence and integrity of the original electronic document at a certain point in time and is tamper-proof.

[0025] Step 103: If the farmer's credit score does not meet the preset conditions, the authenticity of the asset ownership certificate is verified by the evidence documents stored on the key evidence chain. In some embodiments, the preset condition refers to the minimum threshold set for farmers' credit scores in the credit approval process. For example, the rule can be defined as "farmers with a credit score higher than 80 are exempt from asset authenticity verification." Failure to meet the condition means the score is lower than or equal to this threshold. When a farmer's credit score does not meet the preset condition, the backend server invokes the verification service that interacts with the blockchain platform, including: calculating the hash value (e.g., SHA-256) of the received asset ownership certificate and reading the original hash value recorded in the corresponding key evidence on-chain storage file from the blockchain. To ensure the consistency of the asset ownership certificate during on-chain storage and off-chain verification, the backend server normalizes the file before hash value calculation, for example, by uniformly adopting PDF / A format, standardizing and compressing image files, and removing metadata fields. Subsequently, the backend server executes a preset hash algorithm (e.g., SHA-256) on the normalized file byte stream to obtain a standard file hash value. This hash value is stored as a unique fingerprint on the blockchain platform and generates the corresponding key evidence on-chain storage file. The key evidence on-chain not only includes the file hash value, but also the on-chain transaction identifier, block height, block timestamp, hash algorithm type used, and optional Merkle proof fields. When verifying authenticity, the backend server interacts with blockchain nodes based on these notarized documents, reading the hash value and timestamp of the notarized transaction from the chain and comparing it with the normalized hash value of the asset ownership certificate submitted in this application. If they match, the document is determined not to have been tampered with. Furthermore, the backend server can also verify the notarized timestamp to ensure that the notarization occurred before the farmer submitted the application or within a reasonable time window, preventing the illegal reuse of old or expired documents.

[0026] Step 104: If the authenticity verification passes, obtain the corresponding agricultural production process data according to the planting type; and perform structured processing of the agricultural production process data according to the preset data visualization format to form the agricultural production process display data of the target farmer. In some embodiments, agricultural production process data is acquired according to the planting type, including: if the planting type represents open-field planting, then based on the land location information extracted from the asset ownership certificate and the attribute information of the target farmer, standard remote sensing data with ownership marking is queried to obtain the marked plot area and plot area information corresponding to the land location information in the standard remote sensing data. Based on this, real-time remote sensing datasets for multiple time points within the area are retrieved according to the land location information. Specifically, the backend server calculates a bounding rectangle (i.e., the area to which the plot belongs) that can completely cover the plot based on the geographic boundary coordinates of the marked plot area. Then, a query is initiated to a pre-accessed remote sensing data API (e.g., via HTTP request), with request parameters including the geographic area and multiple required time points (e.g., every 15 days within the past 6 months). The remote sensing data platform returns satellite image files that meet the requirements and are arranged in a time series; the backend server receives and stores these files to form a real-time remote sensing dataset. A real-time remote sensing dataset refers to a collection of satellite images collected within a specific time frame and covering the geographical area of ​​a target plot, retrieved by a backend server from an external remote sensing data service provider (such as a satellite data platform). "Real-time" here refers to the latest and most readily available time-series images relative to the entire growth cycle. Each real-time remote sensing data point in the dataset corresponds to a specific time point. Based on the real-time remote sensing dataset, the labeled plot area, and plot area information, agricultural production process data is determined, including: using an instance segmentation model to segment each real-time remote sensing data point, obtaining multiple candidate polygons within each data point; defining the boundary of the labeled plot area as a standard polygon; calculating the similarity between the standard polygon and multiple candidate polygons; and selecting the top N candidate polygons based on similarity for each real-time remote sensing data point; selecting the candidate polygon with the highest frequency among the N candidate polygons corresponding to multiple time points as the target polygon; and cropping the region corresponding to the target polygon in each real-time remote sensing data point to obtain multiple remote sensing image regions corresponding to different time points as agricultural production process data. If the planting type represents greenhouse cultivation, then query the crop growth environment data corresponding to the IoT crop monitoring device identifier within the target historical time period, and generate agricultural production process data. The IoT crop monitoring devices represented by the IoT crop monitoring device identifier include soil environment monitoring sensors, crop growth monitoring sensors, and greenhouse cameras. This includes: if the planting type represents greenhouse cultivation, query the data of soil environment monitoring sensors, crop growth monitoring sensors, and greenhouse cameras within the target historical time period and determine them as agricultural production process data.

[0027] In some embodiments, agricultural production process data refers to remote sensing image areas or IoT monitoring data sets arranged in a time series, reflecting the growth status, environmental conditions, and management operations of crops at different points in time for target farmers. Predefined data visualization formats are predefined formats used to visually display agricultural production process data to users, including charts, line graphs, heat maps, time series animations, dashboard components, etc., which can be generated using front-end visualization libraries (such as ECharts, D3.js, Plotly). Based on this, the back-end server receives the agricultural production process data and maps each data point according to the preset visualization template: a unified time field is added to each data point to generate line graphs, animations, or time series change displays. Cropped images or sensor data are bound to the coordinates of the calibrated plot area to achieve spatial positioning display. Sensor readings are standardized in units, missing or outlier values ​​are handled, and mapped to line graph or heat map color values. The cropped remote sensing images or camera images are stored in chronological order as a multi-frame image sequence, or thumbnails and label information are generated for easy front-end display. Through the above mapping and standardization processes, the raw agricultural production process data is transformed into structured agricultural production process display data. Finally, the organized agricultural production process display data is saved in a target format, such as JSON containing plot IDs, timestamps, image URLs, sensor data fields, etc. The agricultural production process display data for the target farmers is a collection of agricultural production process data that, after structured processing, can be directly used for front-end display. The agricultural production process display data includes time information, plot spatial information, remote sensing image data, IoT sensor data, associated visualization information, and metadata, etc.

[0028] Step 105: Push the agricultural production process display data to the smart terminal in the business hall so that the smart terminal can visualize the display data and initiate interactive inquiries with the target farmers to obtain supplementary information from the target farmers regarding key production nodes, abnormal growth stages, or special agricultural activities. This includes: identifying suspected abnormal features based on the agricultural production process display data; marking the suspected abnormal features to obtain marking results; generating corresponding inquiry items based on the marking results; and sending the inquiry items to the smart terminal so that the smart terminal can present the questions one by one on the interactive interface according to the priority of the inquiry items, guiding the target farmers to provide supplementary information regarding the corresponding key production nodes, abnormal growth stages, or special agricultural activities.

[0029] In some embodiments, the backend server analyzes agricultural production process data to identify suspected anomalies. Anomaly identification can be achieved using quantitative rules or machine learning models. For example, it can calculate the normalized vegetation index for slow-growing or abnormally vigorous areas, perform threshold detection on soil moisture and temperature data, or detect signs of pests and diseases using image recognition models. The backend server performs structured annotation on the identified anomalies (such as slow-growing or vigorous areas, abnormal soil environment, signs of pests and diseases, etc.). Each annotation record includes anomaly code, anomaly type, plot coordinates, time, severity, data source, and description. Annotations are generated through remote sensing image instance segmentation or IoT data threshold judgment, and the severity of the anomaly can be assessed using quantitative rules. All annotation results are stored in a standardized data structure to facilitate the generation of query entries, prioritization, and ensure that agricultural production process information is traceable and verifiable. Each annotation result includes structured fields such as anomaly type, corresponding time point, plot coordinates, and severity. The backend server automatically matches the anomaly type field in the annotation with corresponding pre-set query templates and fills the field values ​​into variables in the template to generate specific query content. The pre-set query templates include various categories such as: a growth anomaly template (e.g., "An anomaly of type 'abnormality' was detected at your plot's {coordinates} at {time}; please confirm if any special management operations were performed?"); an environmental anomaly template (e.g., "Soil moisture is abnormally low; were irrigation adjustments or drainage measures implemented during this period?"); or a pest and disease risk template (e.g., "Image recognition indicates the possible presence of {pest and disease type}; have corresponding symptoms been observed or control measures taken?"). Each query item includes an anomaly description, time point, plot information, and priority. Priority can be quantified based on the severity of the anomaly or credit approval requirements (e.g., level 1-5, with level 5 being the most critical), assigning higher weight to high-risk events such as pests and diseases. The backend server establishes a communication connection with the smart terminals in the business hall, sending query items and agricultural production process display data to the smart terminals. Figure 3The diagram illustrates the interface of an intelligent terminal for generating agricultural production process data according to the present invention. The intelligent terminal is defined as a terminal device with display and network capabilities, such as a tablet computer, desktop touchscreen, or self-service terminal, installed in a business hall. After receiving inquiries, the intelligent terminal presents them one by one on the interactive interface according to their priority, while simultaneously visualizing the agricultural production process data, such as line graphs, heat maps, time-series animations, farmland boundary markings, and remote sensing image displays, enabling farmers to intuitively understand crop growth status and key milestones. During the interaction, farmers input supplementary information, including text descriptions, images, or options, for key production milestones, abnormal growth stages, or special agricultural activities. This supplementary information serves to verify and improve the original agricultural production process data, enhancing its authenticity, completeness, and credibility, and providing a reliable basis for credit approval, risk assessment, and agricultural management. The supplementary information is transmitted back to the backend server via the intelligent terminal and stored in association with the original displayed data, forming a complete closed loop of agricultural production process information. Key production nodes are the time points or operational stages in agricultural production that have a decisive impact on yield or quality, such as sowing, transplanting, topdressing, and irrigation. These are used to supplement core agricultural timing information in the agricultural production process. Abnormal growth stages are time periods in growth monitoring data where suspected anomalies are detected, such as low growth rate or an abnormal decline in leaf area growth rate. These reflect deviations from indicators or abnormal growth intervals identified based on the displayed data. Special agricultural activities refer to operations outside the regular planting process that significantly affect growth or yield, such as temporary pesticide application, emergency irrigation, harvesting, variety change, planting density adjustment, temporary mulching, and remedial measures after hail, strong winds, or waterlogging. These are used to collect information on temporary operations that the backend server cannot automatically detect but that significantly affect crop growth.

[0030] Step 106: Receive supplementary information and associate it with agricultural production process data for storage, generating supporting materials for agricultural production process data.

[0031] In some embodiments, the back-end server receives supplementary information submitted by farmers during interactive inquiries from the smart terminal in the business hall. Upon receiving this supplementary information, the back-end server formats it, such as parsing the text content, extracting the farmer's confirmed operation time, agricultural activity type, or explanation of the anomaly, and generates corresponding data fields. Subsequently, the back-end server associates and stores the supplementary information with the original agricultural production process data. The association method can be based on timestamps, plot coordinates, anomaly feature numbers, production node types, or agricultural activity types. When storing the associated data in the database, the back-end server can generate structured supporting materials for the agricultural production process data. These supporting materials typically include automatically collected remote sensing and sensor data, identified anomaly annotation records, the farmer's supplementary information, interrelated time and plot indexes, and additional images or notes provided by the farmer. The generated supporting materials are used to prove the authenticity and completeness of the agricultural production process, providing credible data support for subsequent agricultural credit approval, risk assessment, agricultural condition verification, and agricultural decision-making. This step combines objective monitoring with subjective verification of agricultural production information, making the agricultural production process auditable, traceable, and more credible, ultimately forming high-quality agricultural production data evidence that can be used in business scenarios.

[0032] In these embodiments, data authenticity and reliability are ensured. Specifically, credit application data, including attribute information, asset certificates, application reasons, and credit limits, is received through farmer terminals. User profiles are queried based on attribute information to obtain credit scores, planting types, IoT device identifiers, and key evidence documents. Asset certificates of farmers with low credit scores are verified for authenticity via blockchain to ensure data credibility. Corresponding agricultural production process data, including remote sensing images or greenhouse IoT monitoring data, is obtained based on planting type, achieving multi-source data integration. Farmer information, production process data, and supplementary explanations are linked and stored to form a complete data chain, which is then structured and visualized. The backend server analyzes and displays the data, identifying key production nodes, abnormal growth stages, and special agricultural activities. Supplementary explanations from farmers are collected through smart terminals to generate supporting materials for agricultural production process data, ensuring data authenticity, completeness, and verifiability, providing a reliable basis for credit approval and agricultural management.

[0033] In some embodiments, to further address the second technical problem described in the background section, namely, "the existing technology lacks high-precision remote sensing data calibration and multi-source agricultural data processing methods for farmers' land plots, making it difficult to accurately extract plot boundaries, identify abnormal states, and quantify disease characteristics for yield prediction," in some embodiments of the present invention, based on land location information extracted from asset ownership certificates and attribute information of the target farmers, standard remote sensing data marked with ownership is queried to obtain the calibrated plot area and plot area information corresponding to the land location information in the standard remote sensing data, including: Based on the attribute information of the target farmers, the standard remote sensing data is queried to obtain the candidate calibration plot area and the corresponding candidate plot area information that match the user identifier of the target farmers; The candidate plot area information is checked for consistency with the land location information. If the check passes, the candidate plot area is determined as the designated plot area, and the candidate plot area information is determined as the plot area information.

[0034] In some embodiments, if the planting type represents open-field planting, the backend server invokes its file parsing module (such as an OCR service) to extract structured land location information from the asset ownership documents submitted by the farmer. This information is typically a cadastral number or plot name (e.g., "Xishangang"). Subsequently, the backend server uses the user identifier (such as an ID card number) obtained from the farmer's attribute information as the query key to perform a spatial query in its maintained standard remote sensing database with ownership annotations. This database stores high-precision benchmark remote sensing images bound to the farmer's ownership information. The query results return one or more candidate labeled plot areas associated with the user identifier and their corresponding candidate plot area information (including geographical boundaries, area, etc.). To ensure the accuracy of the plot information, the backend server then performs a consistency check: comparing the land location information extracted by the OCR with the plot name or number recorded in the candidate plot area information returned by the database. Besides comparisons based on name or number, some embodiments can also employ spatial overlap checks or area error threshold judgments to achieve consistency verification. For example, the overlap rate between the candidate plot boundary and the declared plot boundary can be calculated (using a GIS database or polygon algorithm to calculate the intersection / union area). If the overlap rate exceeds 90%, it is considered consistent. If both are consistent (e.g., the declared "Xishangang" perfectly matches the "Xishangang" recorded in the database), the verification passes. At this point, the backend server officially designates the candidate plot area as the designated plot area used in this approval process and confirms its information as plot area information. Through cross-verification between the ownership database and the farmer's declared information, a reliable data foundation is laid for subsequent remote sensing analysis, effectively preventing the risk of false or misreported plot information. If the verification fails or no candidate plot is found, the process is interrupted and the farmer is notified.

[0035] High-precision benchmark remote sensing imagery typically refers to satellite or UAV imagery with a resolution better than 1 meter per pixel, clearly reflecting plot boundaries and crop distribution. Overlap rate refers to the ratio of the intersection area to the union area formed by the spatial overlay of candidate plot boundaries and declared plot boundaries, used to measure their consistency. OCR (Optical Character Recognition) is a technology that converts text information in image files such as scanned documents and photographs into editable and searchable structured text. Spatial query is a retrieval method based on Geographic Information System (GIS), which can locate matching geographic entities by inputting coordinate ranges, polygon boundaries, or spatial relationships (such as containment or intersection). Candidate calibration plot area refers to the spatial range of one or more plot areas that may belong to a farmer, matched by the backend server based on the farmer's user identifier or attribute information when querying the standard remote sensing database. Candidate plot area information consists of attribute data accompanying the candidate calibration plot area. It typically includes: plot name or number (e.g., "Xishangang Plot"), plot boundary coordinates, plot area, ownership registration number, and possibly historical crop types. It is a textual and structured description of the candidate area, used for subsequent comparison with land location information extracted by OCR. The calibration plot area refers to the officially recognized standard plot range actually operated by the target farmer after consistency verification. It is the final result selected from the candidate calibration plot areas, and all subsequent remote sensing image cropping, crop identification, and growth process analysis are based on this area. Determining the calibration plot area is equivalent to "labeling" the farmer in the standard remote sensing database, ensuring the uniqueness and accuracy of the data. In short, the candidate calibration plot area (spatial range) is combined with the candidate plot area information (attribute data), and after consistency verification, the calibration plot area (the finally confirmed standard range) is obtained.

[0036] In this dataset, each real-time remote sensing data point corresponds to a specific time point. Based on real-time remote sensing datasets, calibrated plot areas, and plot area information, agricultural production process data are determined, including: Step 1: Using an instance segmentation model, perform image segmentation on each real-time remote sensing data in the real-time remote sensing dataset to obtain multiple candidate polygons included in each real-time remote sensing data. In some embodiments, the time point is the specific acquisition date of each remote sensing image, used to identify different stages of crop growth (such as sowing, jointing, and heading stages). Instance segmentation models (e.g., models based on MaskR-CNN or SOLOv2 architectures and trained for agricultural remote sensing images) have been pre-trained and deployed on a backend server. Instance segmentation models are advanced deep learning models for computer vision that can not only identify objects in images but also generate accurate pixel-level contours (i.e., polygons) for each individual object instance. In this invention, this model is specifically trained to identify and segment individual plots of land, possibly farmland, from complex remote sensing images. Figure 2 This invention relates to an image segmentation result in a method for generating agricultural production process data. Based on this, the backend server sequentially submits each image from the real-time remote sensing dataset to an instance segmentation model for inference. For each input image, the instance segmentation model outputs a set containing multiple candidate polygons, each representing the precise boundary of an independent plot of land identified by the instance segmentation model in the image. These candidate polygon sets, organized by time points, are temporarily stored. Here, candidate polygons refer to the geometric boundaries of all potential farmland plots identified by the instance segmentation model after processing a remote sensing image; each polygon consists of a series of vertex coordinates, representing an independent segmented region. As an example, the backend server retrieved satellite images for the four most recent time points (March 1, 2024, April 1, 2024, May 1, 2024, and June 1, 2024) for farmer Zhang San's plot (located in Xishangang). The backend server sends a request to a satellite data platform to obtain images covering the "Xishangang" area for these four dates, forming a real-time remote sensing dataset. The backend server inputs an image from May 1, 2024, into the instance segmentation model. After analysis, the instance segmentation model identifies 15 independent farmland plots on this image and generates 15 precise boundary polygons, which are the 15 candidate polygons corresponding to that time point. The backend server repeats this process, processing all four images in the dataset, ultimately obtaining four sets of candidate polygons, each corresponding to a specific acquisition time point.

[0037] Step 2: Determine the boundary of the calibrated area as a standard polygon, calculate the similarity between the standard polygon and multiple candidate polygons, and select the top N candidate polygons with the highest similarity ranking for each real-time remote sensing data. In some embodiments, the backend server reads the boundary coordinates of the verified labeled area and formally defines it as the standard polygon for this polygon matching. Then, it iterates through each time point in the real-time remote sensing dataset. For a single time point, the server obtains all candidate polygons output by the instance segmentation model at that time point. The backend server calculates pairwise similarity between the standard polygon and each candidate polygon. Key metrics typically include: Intersection over Union (IoU): the ratio of the overlapping area of ​​two polygons to their combined area. This is the most crucial metric for measuring spatial overlap. Shape similarity: for example, by comparing the Fourier descriptors of the two polygons' outlines. Area ratio: the ratio of the areas of two polygons; excessively disparate ratios lower the similarity score. The backend server combines these metrics (e.g., assigning the highest weight to IoU) to calculate a final comprehensive similarity score. Subsequently, the backend server sorts the similarity scores of all candidate polygons at that time point in descending order. Finally, the backend server selects the top N candidate polygons in the sorted list as the "high-potential matching plots" at that time point. In this context, the standard polygon refers to the boundary of a plot of land precisely described using polygon coordinates, based on the designated area, and serves as the benchmark for subsequent matching with candidate polygons. A polygon is the geometric outline on a remote sensing satellite image used to precisely delineate the boundary of a piece of farmland, a greenhouse, or any other independent agricultural land. The top N similarity rankings are selected after calculating the similarity between the standard polygon and all candidate polygons, ranking them from highest to lowest score, and choosing the top N (e.g., the top 3 or top 5) candidates most likely to match the standard polygon. Here, N is a preset integer designed to balance precision and computational complexity. As an example, the backend server already has: a standard polygon: the authoritative boundary coordinates of Zhang San on the "Xishangang" plot; 4 time points, each with approximately 15 candidate polygons; the backend server begins processing the time point May 1, 2024: the backend server retrieves the standard polygon and the 15 candidate polygons for that date; and calculates the overall similarity score (e.g., primarily based on IoU) between the standard polygon and each of these 15 polygons. After calculation, 15 scores are obtained, with the highest being 0.95 and the lowest being 0.1. The server sorts these scores, assuming a preset N of 3, and selects the three candidate polygons with the highest scores (assuming scores of 0.95, 0.83, and 0.79 respectively) as the filtering results for May 1, 2024. The backend server repeats the steps for all other time points, including March 1, 2024.

[0038] Step 3: Select the candidate polygon with the highest frequency from the N candidate polygons corresponding to multiple time points, and use it as the target polygon. Crop the region corresponding to the target polygon in each real-time remote sensing data to obtain multiple remote sensing image regions corresponding to different time points as agricultural production process data.

[0039] In some embodiments, the backend server performs a unified indexing on the N candidate polygons selected at each time point (uniqueness can be represented by polygon vertex coordinates or hash values). The candidate polygon sets at all time points are statistically analyzed, calculating the frequency of each polygon's appearance at different time points. The polygon with the highest frequency is selected as the target polygon. If ties exist, selection can be further based on the rule of the polygon with the closest area to the standard polygon or the largest IoU with the standard polygon. Based on this, for each real-time remote sensing image, the backend server crops the image according to the vertex coordinates of the target polygon. The output image size can be consistent with the original resolution or uniformly scaled according to analysis needs. These cropped image fragments are stored in chronological order to form a complete agricultural production process dataset. The target polygon refers to the polygon with the highest statistical frequency among the N candidate polygon sets at multiple time points, i.e., the farmland boundary most likely corresponding to the standard polygon. The cropped region refers to extracting the region corresponding to the target polygon in each real-time remote sensing image to form a separate image fragment for subsequent crop identification or growth analysis. Remote sensing image regions refer to the cropped sub-images. These regions have had irrelevant environmental information removed, focusing only on the target farmland itself, and are arranged in a time series. As an example, suppose there are four time points for farmer Zhang San's plot of land. For each time point, N = three candidate polygons are selected: Time point 1: A1, A2, A3; Time point 2: A2, B1, B2; Time point 3: A2, C1, C2; Time point 4: A2, D1, D2. Among these, A2 appears 4 times, the highest frequency, therefore A2 is selected as the target polygon. The backend server uses A2 to crop the corresponding region from each remote sensing image, obtaining four cropped images that form the agricultural production process dataset.

[0040] Among them, the IoT crop monitoring equipment identifier represents IoT crop monitoring equipment including soil environment monitoring sensors, crop growth monitoring sensors, and greenhouse cameras; and If the planting type represents greenhouse cultivation, then query the crop growth environment data corresponding to the IoT crop monitoring device identifier within the target historical time period, and generate agricultural production process data, including: If the planting type represents greenhouse planting, then query the data from soil environmental monitoring sensors, crop growth monitoring sensors, and greenhouse cameras within the target historical time period and identify them as agricultural production process data.

[0041] In some embodiments, IoT crop monitoring equipment refers to a collection of intelligent hardware devices deployed inside agricultural greenhouses to automatically collect various environmental and crop growth data. Soil environment monitoring sensors are responsible for monitoring the environmental conditions around crop roots, typically including soil temperature and humidity sensors, soil pH sensors, and soil conductivity sensors. Crop growth monitoring sensors are responsible for directly monitoring the physiological state of the crop itself or the canopy environment, such as leaf temperature and humidity sensors, canopy light sensors, and carbon dioxide concentration sensors. In some high-precision agriculture, stem flow sensors (measuring crop transpiration) may also be included. Greenhouse cameras are fixed installations inside the greenhouse, used to periodically capture visible light or near-infrared images of crops, providing the most intuitive visual evidence of crop morphology, color, and overall growth. Based on this, after determining the planting type as "greenhouse cultivation," the backend server retrieves the list of IoT crop monitoring device identifiers pre-configured by the farmer from their user profile. According to the farmer's IoT crop monitoring device identifiers, the backend server initiates a query request to the IoT data management platform. The query criteria include: target time period (e.g., the past quarter or year), device type (e.g., soil sensor, crop growth sensor, camera), and device identifier. The IoT data management platform returns data including: numerical environmental and growth parameter data tables (with timestamps), and image or video sequences (with timestamps and camera IDs). Subsequently, the backend server organizes and formats the retrieved data, arranging it in chronological order. The organized data set serves as the farmer's greenhouse agricultural production process data. The IoT data management platform is a centralized software system located between IoT devices (sensors, cameras, etc.) and end-user applications (e.g., the credit approval system in this patent). It is responsible for the unified access, management, and monitoring of massive, heterogeneous IoT devices, and for collecting, storing, processing, and analyzing the data generated by these devices, ultimately providing it to upper-layer applications in a standardized manner.

[0042] The method for generating agricultural production process data according to the present invention further includes: Step 1: If the planting type represents greenhouse planting, then extract the greenhouse images from the greenhouse camera collection according to the preset image extraction ratio to obtain the extracted greenhouse image set; according to the preset M time intervals and the collection timestamp of each greenhouse image, divide the extracted greenhouse image set into M greenhouse image groups, and the M greenhouse image groups correspond one-to-one with the M time intervals. In some embodiments, the preset image extraction ratio is a pre-defined ratio (e.g., 10%) used to downsample the original image set. The extraction ratio can be set by the backend server according to the application scenario. The greenhouse image set refers to the collection of all images automatically captured and uploaded to the server by greenhouse cameras within the target historical time period. Since cameras may capture images periodically (e.g., hourly), the data volume is enormous. If the planting type represents greenhouse cultivation, the backend server retrieves the complete greenhouse image set of the target farmer within the target historical time period from storage. The backend server samples according to the preset image extraction ratio. For example, if the ratio is 10%, one image is extracted from every ten images. All the extracted images are combined into a new extracted greenhouse image set. The extracted greenhouse image set refers to the subset obtained by filtering from the original greenhouse image set according to the image extraction ratio. Based on this, the backend server creates M empty greenhouse image groups according to M preset time intervals (e.g., [start date, day 30], (day 30, day 60], (day 60, day 90], (day 90, end date]). The time intervals can be divided by a fixed number of days (e.g., every 30 days) or by combining the crop growth cycle (e.g., sowing period, seedling stage, heading stage, maturity stage). The backend server iterates through and extracts each image from the greenhouse image set, reading its acquisition timestamp. Based on the timestamp, it determines which time interval the image belongs to and places it into the corresponding greenhouse image group. After the iteration is complete, all images are assigned to the M time intervals. The data is distributed across M corresponding greenhouse image sets. The acquisition timestamp is the precise capture time recorded in each image file. The M time intervals divide the entire target historical period (e.g., a growing season) into M consecutive, non-overlapping stages. For example, a 120-day growing season can be divided into M=4 intervals, each lasting 30 days. A greenhouse image set refers to the collection of images formed by dividing the extracted images into corresponding time intervals according to their timestamps. Each time interval corresponds to one image set, ensuring a uniform temporal distribution of data. The purpose of image extraction and grouping is to reduce the computational burden on subsequent recognition and prediction models while ensuring data coverage across different stages, all while maintaining data representativeness.

[0043] Step 2: Identify the extracted greenhouse image set to obtain the greenhouse crop category. Based on the greenhouse crop category, query the pre-configured disease feature database to obtain at least one typical disease feature corresponding to the greenhouse crop category and the disease stage corresponding to each typical disease feature. In some embodiments, the backend server extracts greenhouse images one by one and inputs them into a pre-trained crop recognition model (e.g., a hybrid network of CNN or Transformer+CNN). Output: The output identifies the crop category label corresponding to each image, such as "tomato," "cucumber," or "strawberry." Since greenhouses typically grow a single crop, the backend server performs statistical analysis (e.g., voting or confidence-weighted summation) on all recognition results to determine the crop category of the greenhouse during the target historical time period. The pre-trained crop recognition model refers to an artificial intelligence model that has been trained and optimized using a large amount of labeled crop image data before being deployed to the backend server. Its core function is to automatically identify the crop category (e.g., rice, wheat, corn, tomato, cucumber, etc.) contained in the input images and output category labels and confidence scores. The disease feature library refers to a pre-built database that stores the correspondence between crops and diseases, including image features, textual descriptions, disease conditions, and corresponding disease stages of typical diseases. After obtaining the crop category in the greenhouse, the backend server uses the identified crop category as the search condition to retrieve the typical disease characteristics and disease stages corresponding to that crop from the disease characteristic database. One crop category typically corresponds to multiple typical diseases, and each disease corresponds to different disease stages. Typical disease characteristics refer to the most representative, easily identifiable, or highly damaging disease symptoms for a particular crop. Disease stages refer to the characteristics of the disease at different growth stages of the crop (or the early / middle / late stages of the disease process), used to guide disease prediction and control.

[0044] Step 3: Based on the disease stage corresponding to each typical disease characteristic, assign each typical disease characteristic to one or more time intervals; In some embodiments, the backend server maintains a mapping rule table that associates general agricultural "disease stage" terms with specific, quantifiable "time interval" definitions. For each typical disease feature and its disease stage retrieved from the disease feature database, the backend server's scheduling module performs the following operations: reads the disease stage (e.g., the disease stage of "strawberry powdery mildew" is "mid-to-late growth"). Based on the mapping rule table, it converts the agricultural term "mid-to-late growth" into a specific time interval number defined in step one. For example, the rule might define "mid-to-late growth" as corresponding to time intervals 3 and 4. It then performs an allocation operation, formally associating the "strawberry powdery mildew" disease feature with time intervals 3 and 4. After allocation, each time interval has one (or more) lists of typical disease features that require focused image analysis and screening within that time period. Here, the time intervals are M consecutive, specific time periods (e.g., specific date ranges) pre-defined by the backend server in step one, dividing the entire historical growing season.

[0045] Step 4: For each greenhouse image group, extract the image features of each greenhouse image and calculate the average image feature within the group. Based on the Euclidean distance between the average image feature and each image feature, divide each greenhouse image group into an abnormal greenhouse image group and a normal greenhouse image group. Determine the average abnormal image feature corresponding to the abnormal greenhouse image group and calculate its similarity with each corresponding typical disease feature. If the similarity is greater than or equal to a preset similarity threshold, then the corresponding typical disease feature is identified as a potential disease feature. In some embodiments, for each greenhouse image in each greenhouse image group, the backend server processes it using a pre-trained image feature extraction model (such as a convolutional neural network (CNN) or a deep feature extraction network), outputting the image feature vector. Image features refer to high-dimensional vectors calculated by the model, used to characterize the visual information of the image, such as color, texture, shape, and disease features, for subsequent similarity comparison or cluster analysis. For each greenhouse image group, the feature vectors of all images in the group are averaged to obtain the average image feature of the group, representing the overall state of the greenhouse within that time interval. Based on this, the Euclidean distance between the feature vectors of each image in the group and the average image feature is calculated to measure the degree of deviation of the image from the overall state. According to a preset threshold, images with large deviations are classified as abnormal images. Further, the abnormality of the entire greenhouse image group is statistically analyzed, dividing the greenhouse image group into abnormal greenhouse image group and normal greenhouse image group. For the abnormal greenhouse image group, the average feature of all abnormal images in the group is calculated to obtain the average abnormal image feature of the group. Subsequently, the average abnormal image features are compared with the typical disease features of the corresponding time interval in step three using similarity calculations (such as cosine similarity, reciprocal Euclidean distance, etc.). If the similarity is greater than or equal to a preset similarity threshold, the typical disease feature is marked as a potential disease feature. Here, an abnormal greenhouse image group refers to a set of images within a given time interval that exhibits overall abnormal behavior and deviates significantly from the average state, potentially indicating disease or other anomalies. The average abnormal image feature is the mean of all abnormal image features within the abnormal image group, representing the overall characteristics of the group's anomalies. Potential disease features are disease features identified through similarity matching within a specific time interval and image group, used for subsequent disease prediction and prevention. The preset similarity threshold is a pre-configured value within the backend server, used to determine whether image features and disease features are sufficiently matched, often determined based on historical data or expert experience. Through the above-mentioned intra-group anomaly detection based on Euclidean distance, the backend server can distinguish local anomalies from normal fluctuations, ensuring that subsequent disease similarity identification is performed only on abnormal groups, improving computational efficiency and reducing false alarms. Meanwhile, the average abnormal image features, as a global representation of abnormal states, can effectively suppress the influence of single-frame noise on disease matching and improve the stability and accuracy of potential disease feature identification.

[0046] Step 5: Generate yield predictions based on potential disease characteristics and crop growth index data sequences measured by crop growth monitoring sensors. Step 6: Add the yield forecast and the potential disease characteristics corresponding to each time interval to the agricultural production process data.

[0047] In some embodiments, the backend server combines potential disease characteristics with time series data of growth indicators collected by crop growth monitoring sensors to form a fused feature vector sequence, which can be achieved using vector concatenation or a multi-branch feature fusion network. The time series data of growth indicators are crop growth-related data collected chronologically within the target historical time period, including indicators such as leaf area index, plant height, canopy greenness, and leaf surface temperature and humidity. The backend server inputs the fused feature vector sequence into a yield prediction model, which is a time-series prediction network, such as a long short-term memory network or a time-series convolutional network. The yield prediction model, through forward computation, combines crop growth status, potential disease information, and time series variation patterns to output a corrected yield prediction value. The yield prediction value may include yield per unit area or total yield, reflecting the future crop yield level of the target plot or greenhouse. The introduction of potential disease characteristics enables the model to correct for disease risks that may affect crop growth, improving the accuracy of yield prediction. After generating the yield prediction value, the backend server also compares the prediction result with a historical yield deviation model to verify the reliability of the prediction. When the predicted yield decline exceeds a preset threshold, the system can automatically trigger a risk alert to guide agricultural management (such as increasing irrigation and early prevention) or assist in credit risk control. In this way, the yield forecast not only serves as a static assessment but also provides dynamic guidance. Ultimately, the yield forecast and potential disease characteristics are incorporated into the agricultural production process data in a structured form (JSON object or database table). Integrating the predicted yield and potential disease information into the agricultural production process data achieves information closure and structured storage, providing a complete and quantifiable data foundation for agricultural management and credit assessment. The agricultural production process data includes multiple records, each corresponding to a time interval or a plot / greenhouse, and can nest multiple fields, including the time interval, a list of potential disease characteristics, the predicted yield, and associated raw sensor data and image data.

[0048] These embodiments enable accurate monitoring of crop growth processes, early disease risk quantification and warning, and precise yield prediction. Specifically, by extracting land location information and farmer attributes from asset ownership documents, candidate plots are queried in a standard remote sensing database, and consistency verification is performed through name or number comparison and spatial overlay calculation to achieve high-precision and verifiable plot labeling, ensuring accurate and reliable boundaries at different time points. For open-field cultivation, candidate polygons in real-time remote sensing data are identified through an instance segmentation model, and similarity screening is performed with the labeled plot polygons to generate agricultural production process data, achieving automated and time-series data acquisition. For greenhouse cultivation, this application divides historical data collected by soil environment, crop growth sensors, and cameras into time intervals and performs image extraction and organization to form a structured dataset, providing a foundation for subsequent analysis. By extracting greenhouse image features and matching typical disease features, automatic identification and quantification of abnormal images and potential disease features are achieved, improving the accuracy of disease monitoring. Ultimately, by combining potential disease characteristics with time series of crop growth indicators, a yield prediction model is generated and the predicted values ​​are integrated into agricultural production data to achieve more accurate and dynamic yield prediction, providing reliable data support for agricultural management and credit risk control.

[0049] The above description is merely a selection of preferred embodiments of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to specific combinations of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this invention.

Claims

1. A method for generating agricultural production process data, characterized in that, include: The system receives credit application data from target farmers through farmer terminals. The credit application data includes the target farmers' attribute information, asset ownership documents, and the reasons and amount of the credit application. Based on the attribute information, query the pre-built user profile to obtain the target farmer's farmer integrity score, planting type, the identifier of the IoT crop monitoring device pre-configured by the farmer, and the evidence files stored on the key evidence chain; If the farmer's integrity score does not meet the preset conditions, the authenticity of the asset ownership certificate will be verified through the evidence documents stored on the key evidence chain. If the authenticity verification passes, the corresponding agricultural production process data is obtained according to the planting type; the agricultural production process data is then structured according to a preset data visualization format to form the agricultural production process display data of the target farmer. The agricultural production process display data is pushed to the smart terminal in the business hall so that the smart terminal can visualize the display data and initiate interactive inquiries to the target farmers to obtain supplementary information from the target farmers on key production nodes, abnormal growth stages or special agricultural activities. The system receives the supplementary information and associates it with the agricultural production process data for storage, generating supporting materials for the agricultural production process data.

2. The method for generating agricultural production process data according to claim 1, characterized in that, The step of obtaining corresponding agricultural production process data based on the planting type includes: If the planting type represents open-field planting, then based on the land location information extracted from the asset ownership certificate and the attribute information of the target farmer, the standard remote sensing data marked with ownership is queried to obtain the marked plot area and plot area information corresponding to the land location information in the standard remote sensing data; and based on the land location information, real-time remote sensing datasets of multiple time points within the area are pulled, and based on the real-time remote sensing datasets, marked plot areas and plot area information, agricultural production process data are determined; If the planting type represents greenhouse planting, then query the crop growth environment data corresponding to the IoT crop monitoring device identifier within the target historical time period, and generate agricultural production process data.

3. The method for generating agricultural production process data according to claim 2, characterized in that, Each real-time remote sensing data point in the real-time remote sensing dataset corresponds to a specific time point. The process of determining agricultural production process data based on real-time remote sensing datasets, calibrated plot areas, and plot area information includes: Using an instance segmentation model, image segmentation is performed on each real-time remote sensing data in the real-time remote sensing dataset to obtain multiple candidate polygons included in each real-time remote sensing data. The boundary of the calibrated area is defined as a standard polygon. The standard polygon is compared with the multiple candidate polygons. The top N candidate polygons with the highest similarity ranking for each real-time remote sensing data are selected. The candidate polygon with the highest frequency among N candidate polygons corresponding to multiple time points is selected as the target polygon. The region corresponding to the target polygon in each real-time remote sensing data is cropped to obtain multiple remote sensing image regions corresponding to different time points as agricultural production process data.

4. The method for generating agricultural production process data according to claim 3, characterized in that, The IoT crop monitoring equipment identified by the IoT crop monitoring equipment identifier includes soil environment monitoring sensors, crop growth monitoring sensors, and greenhouse cameras; and If the planting type represents greenhouse cultivation, then the system queries the crop growth environment data corresponding to the IoT crop monitoring device identifier within the target historical time period and generates agricultural production process data, including: If the planting type represents greenhouse planting, then query the data from soil environmental monitoring sensors, crop growth monitoring sensors, and greenhouse cameras within the target historical time period and identify them as agricultural production process data.

5. The method for generating agricultural production process data according to claim 4, characterized in that, Also includes: If the planting type represents greenhouse planting, then the greenhouse images collected by the greenhouse camera are extracted according to a preset image extraction ratio to obtain the extracted greenhouse image set. According to the preset M time intervals and the acquisition timestamp of each greenhouse image, the extracted greenhouse image set is divided into M greenhouse image groups, and the M greenhouse image groups correspond one-to-one with the M time intervals; The extracted greenhouse image set is identified to obtain the greenhouse crop category. Based on the greenhouse crop category, a pre-configured disease feature database is queried to obtain at least one typical disease feature corresponding to the greenhouse crop category and the disease stage corresponding to each typical disease feature. Based on the disease stage corresponding to each typical disease characteristic, each typical disease characteristic is assigned to one or more time intervals.

6. The method for generating agricultural production process data according to claim 5, characterized in that, Also includes: For each greenhouse image group, the image features of each greenhouse image are extracted, and the average image features within the group are calculated. Based on the Euclidean distance between the average image features and each image feature, each greenhouse image group is divided into an abnormal greenhouse image group and a normal greenhouse image group. Determine the average abnormal image features corresponding to the abnormal greenhouse image group, and calculate the similarity with each corresponding typical disease feature; if the similarity is greater than or equal to the preset similarity threshold, the corresponding typical disease feature is determined as a potential disease feature. Based on the potential disease characteristics and the crop growth index data sequence measured by the crop growth monitoring sensor, a yield prediction value is generated; The predicted yield values ​​and the potential disease characteristics corresponding to each time interval are added to the agricultural production process data.

7. The method for generating agricultural production process data according to claim 6, characterized in that, The step of querying standard remote sensing data marked with ownership information based on the land location information extracted from the asset ownership certificate and the attribute information of the target farmer to obtain the marked plot area and plot area information corresponding to the land location information in the standard remote sensing data includes: Based on the attribute information of the target farmers, the standard remote sensing data is queried to obtain candidate calibration plot areas and corresponding candidate plot area information that match the user identifier of the target farmers; The candidate plot area information is checked for consistency with the land location information. If the check passes, the candidate marked plot area is determined as the marked plot area, and the candidate plot area information is determined as the plot area information.

8. The method for generating agricultural production process data according to claim 7, characterized in that, The process involves pushing the agricultural production process data to the smart terminal in the business hall, enabling the smart terminal to visualize the data and initiate interactive inquiries with the target farmers. This allows the target farmers to provide supplementary information regarding key production nodes, abnormal growth stages, or special agricultural activities, including: Based on the agricultural production process data, identify suspected abnormal features; annotate the suspected abnormal features to obtain annotation results; and generate corresponding query entries based on the annotation results. The query items are sent to the smart terminal, so that the smart terminal presents the questions one by one on the interactive interface according to the priority of the query items, and guides the target farmers to provide supplementary explanations for the corresponding key production nodes, abnormal growth stages or special agricultural activities, and obtain supplementary explanation information.

Images