Farmland multi-source data quality self-calibration method and system

Abstract

The application provides a farmland multi-source data quality self-calibration method and system. The method comprises the following steps: acquiring multi-source farmland original data of a target farmland, and generating a farmland space-time calibration unit; mapping the multi-source farmland original data to the farmland space-time calibration unit, extracting quality characteristics, and generating a data source quality portrait; constructing a farmland quality constraint graph based on the farmland space-time calibration unit, agronomic process constraints and the data source quality portrait; determining a dynamic trusted anchor point according to the data source quality portrait and the farmland quality constraint graph, and generating a corresponding self-calibration reference state; identifying a data quality defect type, attributing a deviation source to the data quality defect type, and generating a deviation attribution result; generating a data source self-calibration parameter according to the deviation attribution result, correcting the multi-source farmland original data by using the data source self-calibration parameter, and generating a trusted farmland data view. The application can improve data quality identification accuracy, enhance deviation attribution reliability, and improve calibration update efficiency.

Description

Technical Field

[0001] This application relates to the field of smart agriculture data processing technology, and in particular to a method and system for self-calibrating the quality of multi-source farmland data. Background Technology

[0002] With the development of the Internet of Things, remote sensing monitoring, intelligent agricultural machinery, and agricultural big data technologies, continuous acquisition of multi-source data, including soil moisture, field weather, crop growth, irrigation and fertilization, agricultural machinery operations, and manual sampling, is possible during farmland production. This data has become a crucial foundation for precision irrigation, variable-rate fertilization, pest and disease early warning, and yield prediction. Existing technologies typically clean and comprehensively analyze farmland data by standardizing data formats, setting anomaly thresholds, imputing missing values, or employing multi-source fusion models to improve the accuracy of crop condition identification and agricultural decision-making.

[0003] However, farmland data sources are complex, with significant differences in collection frequency, spatial scale, observed objects, and data semantics among different data sources. Issues such as sensor drift, communication packet loss, remote sensing cloud shadow interference, agricultural machinery trajectory deviation, and lag in manual recording can easily lead to unstable data quality. Existing methods often rely on single-source thresholds or fixed reference data to identify anomalies, making it difficult to distinguish between changes in farmland status caused by real agronomic events and data quality defects. At the same time, existing data cleaning methods typically only remove or compensate for outlier data, lacking mechanisms for quality profiling based on data sources, constructing reliable references, attributing bias sources, and updating calibration parameters. This prevents continuous self-calibration of data quality defects, meaning subsequent agricultural decisions may still be influenced by low-reliability data. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method and system for self-calibration of farmland multi-source data quality, in order to solve the problems of difficulty in identifying the quality of multi-source data, difficulty in attributing the source of deviation, and difficulty in updating calibration parameters in the prior art.

[0005] A first aspect of this application provides a method for self-calibrating the quality of multi-source farmland data, comprising: acquiring multi-source raw farmland data of a target farmland, and generating a farmland spatiotemporal calibration unit based on plot spatial information, agronomic management information, and acquisition time information; mapping the multi-source raw farmland data to the farmland spatiotemporal calibration unit, extracting quality features, and generating a data source quality profile; constructing a farmland quality constraint map characterizing observation correlation, state evolution, and quality transmission relationships based on the farmland spatiotemporal calibration unit, agronomic process constraints, and the data source quality profile; and determining dynamic reliability based on the data source quality profile and the farmland quality constraint map. The system identifies anchor points and generates corresponding self-calibration reference states. Based on the cross-source consistency residuals between observations from each data source and the self-calibration reference states, as well as the process constraint residuals between observations from each data source and agronomic process constraints, it identifies data quality defect types. It then uses the farmland quality constraint map to attribute the sources of these defects, generating bias attribution results. Based on the bias attribution results, it generates data source calibration parameters, uses these parameters to correct the original multi-source farmland data, generates a reliable farmland data view, and updates the data source quality profile, farmland quality constraint map, and data source calibration parameters based on subsequent feedback.

[0006] A second aspect of this application provides a self-calibration system for the quality of multi-source farmland data, comprising: an acquisition module for acquiring multi-source raw farmland data of a target farmland and generating a farmland spatiotemporal calibration unit based on plot spatial information, agronomic management information, and collection time information; an extraction module for mapping the multi-source raw farmland data to the farmland spatiotemporal calibration unit, extracting quality features, and generating a data source quality profile; a construction module for constructing a farmland quality constraint map characterizing observation correlation, state evolution, and quality transmission relationships based on the farmland spatiotemporal calibration unit, agronomic process constraints, and the data source quality profile; and a determination module for determining, based on the data source quality profile and the farmland quality constraint map. The system comprises: a dynamic and reliable anchor point, generating a corresponding self-calibration reference state; an identification module, used to identify data quality defect types based on the cross-source consistency residuals between observations from each data source and the self-calibration reference state, and the process constraint residuals between observations from each data source and agronomic process constraints; a generation module, used to attribute the source of deviations to the data quality defect types based on the farmland quality constraint map, generating deviation attribution results; and a correction module, used to generate data source calibration parameters based on the deviation attribution results, use the data source calibration parameters to correct the original multi-source farmland data, generate a reliable farmland data view, and update the data source quality profile, farmland quality constraint map, and data source calibration parameters based on subsequent feedback.

[0007] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: By acquiring multi-source raw farmland data of the target farmland and generating farmland spatiotemporal calibration units based on plot spatial information, agronomic management information, and collection time information, the multi-source raw farmland data is mapped to the farmland spatiotemporal calibration units. Quality features are extracted to generate data source quality profiles. Based on the farmland spatiotemporal calibration units, agronomic process constraints, and data source quality profiles, a farmland quality constraint map representing observation correlations, state evolution, and quality transmission relationships is constructed. Dynamic reliable anchor points are determined based on the data source quality profiles and farmland quality constraint maps, and corresponding self-calibration reference states are generated. Data quality defect types are identified based on the cross-source consistency residuals between observations from each data source and the self-calibration reference states, as well as the process constraint residuals between observations from each data source and agronomic process constraints. Based on the farmland quality constraint maps, the sources of deviation for data quality defect types are attributed, generating deviation attribution results. Data source calibration parameters are generated based on the deviation attribution results. The multi-source raw farmland data is corrected using the data source calibration parameters to generate a reliable farmland data view. The data source quality profiles, farmland quality constraint maps, and data source calibration parameters are updated based on subsequent feedback. This application can improve the accuracy of data quality identification, enhance the reliability of deviation attribution, and improve the efficiency of calibration update. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the self-calibration method for farmland multi-source data quality provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the farmland multi-source data quality self-calibration system provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0010] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0011] With the development of smart agriculture, agricultural IoT, remote sensing monitoring, and intelligent agricultural machinery technologies, diverse and complex agricultural data has been generated during the agricultural production process. Existing technologies typically acquire agricultural status data through sensor collection, remote sensing inversion, agricultural machinery terminal recording, and manual input. Data is then used for precision irrigation, variable-rate fertilization, pest and disease early warning, and yield prediction through methods such as format conversion, anomaly threshold judgment, missing value imputation, and multi-source data fusion.

[0012] However, existing farmland data processing methods primarily focus on data aggregation and application analysis, lacking the ability to dynamically identify the quality status of the data sources themselves. Because different data sources differ in collection frequency, spatial scale, observed objects, and data semantics, and because farmland conditions are affected by rainfall, irrigation, fertilization, agricultural machinery operations, and crop growth stages, existing anomaly detection methods based on single-source thresholds or fixed reference data struggle to accurately distinguish between genuine agronomical changes and data quality defects. Furthermore, existing data cleaning methods typically only remove or compensate for outliers, failing to further determine the source of bias or generate continuously updated data source calibration parameters.

[0013] To address the aforementioned issues, this application provides a method for self-calibrating the quality of multi-source farmland data. The method involves acquiring multi-source raw farmland data of the target farmland and generating farmland spatiotemporal calibration units based on plot spatial information, agronomic management information, and data collection time information. The multi-source raw farmland data is then mapped to these spatiotemporal calibration units, quality features are extracted, and a data source quality profile is generated. A farmland quality constraint map is constructed based on the farmland spatiotemporal calibration units, agronomic process constraints, and the data source quality profile. Furthermore, dynamic reliable anchor points are determined based on the data source quality profile and the farmland quality constraint map to generate a self-calibration reference state. Finally, cross-source consistency residuals and process constraint residuals are combined to identify data quality defect types.

[0014] Building upon this foundation, this application further attributes the sources of deviations to data quality defect types based on farmland quality constraint maps, generates deviation attribution results, and generates data source calibration parameters based on these results. These parameters are then used to correct the original multi-source farmland data, generating a reliable farmland data view. Simultaneously, the data source quality profile, farmland quality constraint map, and data source calibration parameters are updated based on subsequent feedback. Therefore, this application improves the accuracy of identifying multi-source farmland data quality, enhances the reliability of deviation attribution, and improves the continuous update efficiency of data source calibration parameters, thereby increasing the credibility and stability of the data upon which subsequent agricultural decisions are based.

[0015] The technical solution of this application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0016] Figure 1This is a flowchart illustrating the self-calibration method for farmland multi-source data quality provided in this application embodiment. For example... Figure 1 As shown, the method may specifically include: S101, acquire multi-source farmland raw data of the target farmland, and generate farmland spatiotemporal calibration units based on plot spatial information, agronomic management information and collection time information; S102, map multi-source farmland raw data to farmland spatiotemporal calibration unit, extract quality features, and generate data source quality profile; S103, based on farmland spatiotemporal calibration units, agronomic process constraints, and data source quality profiles, constructs a farmland quality constraint map characterizing the relationship between observation correlation, state evolution, and quality transmission; S104, determine the dynamic reliable anchor point based on the data source quality profile and farmland quality constraint map, and generate the corresponding self-calibration reference state; S105. Identify the types of data quality defects based on the cross-source consistency residuals between the observations from each data source and the self-calibration reference state, as well as the process constraint residuals between the observations from each data source and the agronomic process constraints. S106, Attribute the sources of deviations to the types of data quality defects based on the farmland quality constraint map, and generate deviation attribution results; S107: Generate data source calibration parameters based on deviation attribution results, use data source calibration parameters to correct multi-source farmland original data, generate a reliable farmland data view, and update the data source quality profile, farmland quality constraint map and data source calibration parameters based on subsequent feedback.

[0017] In some embodiments, multi-source farmland raw data of the target farmland is acquired, and a farmland spatiotemporal calibration unit is generated based on plot spatial information, agronomic management information, and acquisition time information, including: According to the data access protocol of the target farmland, the raw farmland data from multiple sources carrying the collected metadata is received, and the data source identifier, collection location, collection time and observation object are parsed from the collected metadata. Based on the spatial information of the land parcel, the target farmland is divided into farmland objects to generate spatial calibration objects; Based on agronomic management information, configure agronomic status identifiers for spatial calibration objects, and determine the corresponding time calibration window based on the collection time information; By binding spatial calibration objects, agronomic status identifiers, and time calibration windows, a farmland spatiotemporal calibration unit with data carrying relationship is generated.

[0018] Specifically, in application scenarios where target farmland is equipped with ground-based soil moisture collection terminals, field weather stations, UAV image acquisition equipment, agricultural machinery operation terminals, and agricultural management platforms, the data access service pre-configures a unified data access protocol for different data sources. The data access protocol includes data source registration rules, message field structures, coordinate references, time references, observation object codes, and transmission verification rules. When a ground-based soil moisture collection terminal reports soil moisture content, soil temperature, and electrical conductivity data, the data access service identifies the corresponding data source identifier based on the terminal number, parses the collection location based on the terminal installation coordinates, parses the collection time based on the message timestamp, and determines whether the observed object is soil moisture state or soil physicochemical state based on the variable code.

[0019] When UAV image acquisition equipment uploads orthophotos and vegetation index inversion results, the data access service analyzes the corresponding data source identifier, image coverage, imaging time, and crop growth observation object based on the flight path mission number, image exterior orientation elements, and inversion mission identifier. When agricultural machinery operation terminals upload sowing, fertilization, or irrigation operation trajectories, the data access service analyzes the agricultural machinery equipment number, trajectory point coordinates, operation start and end times, and operation object, ensuring that raw data from different sources all carry acquisition metadata that can be used for spatiotemporal aggregation.

[0020] After data access is completed, the system uses the spatial information of the target farmland to perform object-oriented segmentation of the farmland. The spatial information of the plots can consist of plot boundaries, field ridge lines, road boundaries, irrigation canal systems, soil type zoning, and historical yield zoning. The system first uses the plot boundaries as the basic constraint, eliminating roads, canals, and non-planted areas, and then combines irrigation zoning and soil zoning to spatially segment the target farmland, obtaining spatially calibrated objects with clear management implications.

[0021] For example, for a 120-mu (approximately 8 hectares) rice experimental field, the system first divides the field into four planting areas according to the field boundaries, then further divides each planting area into several irrigation management areas based on the control range of the irrigation branch canals, and finally generates spatial calibration objects by combining soil texture differences. Each spatial calibration object is configured with a spatial object identifier, boundary geometric information, center coordinates, area attributes, and adjacent object relationships to carry the multi-source observation data that will be subsequently mapped in.

[0022] Furthermore, the system configures agronomic status identifiers for spatial calibration objects based on agronomic management information. This agronomic management information includes crop type, variety information, planting date, growth stage, irrigation plan, fertilization plan, plant protection plan, and agricultural machinery operation records. The system determines the current crop growth stage of the spatial calibration object based on the planting date and accumulated temperature progress, and combines this with recent agricultural events to determine the agronomic status identifier. For example, if a spatial calibration object is planted with rice and is currently in the tillering stage, and irrigation was completed that morning, then the spatial calibration object will be configured with an agronomic status identifier that includes crop type, growth stage, and post-irrigation status. If another spatial calibration object is in the jointing and booting stage and has plant protection operation records, then the corresponding agronomic status identifier reflects the crop growth status and plant protection management status. In this way, the system can transform simple spatial objects into calibration objects with agronomic semantics.

[0023] For data acquisition time information, the system determines the time calibration window based on the acquisition frequency and agronomic response lag of different data sources. For ground sensor data reported on a minute-by-minute basis, the system can form a time calibration window of 30 minutes or 1 hour; for daily remote sensing inversion data, the system can form a daily-scale time calibration window by combining imaging time and effective coverage area; for agricultural machinery operation data, the system generates an event-driven time calibration window based on the start and end times of the operation and the duration of its post-operation impact. If an irrigation event occurs between 9:00 and 10:00 AM, the system can use this time period and 2 hours after irrigation as the same moisture response calibration window, so that the increase in soil moisture after irrigation will not be mistaken for an acquisition anomaly.

[0024] Finally, the system binds spatial calibration objects, agronomic status identifiers, and temporal calibration windows to generate farmland spatiotemporal calibration units. Each farmland spatiotemporal calibration unit has a unique unit identifier and records its corresponding spatial range, agronomic status, temporal range, and the types of data sources it can receive. When ground sensors, UAV imagery, agricultural machinery trajectories, and manual field inspection records enter the system, they are all matched to the corresponding farmland spatiotemporal calibration units based on the acquisition location, acquisition time, and observation object, thus forming a data carrying relationship. As a result, multi-source farmland raw data no longer exists as isolated data streams, but is organized into a unified calibration object with spatial boundaries, agronomic semantics, and temporal constraints, providing a consistent spatiotemporal basis for subsequent quality feature extraction, reliable anchor point determination, and data self-calibration.

[0025] Through the above processing, the system can solve the problems of inconsistent collection locations, inconsistent time scales, and inconsistent agronomic semantics of multi-source farmland data. This allows subsequent data quality judgment to no longer rely on a single data source or fixed threshold, but to be uniformly carried and aligned based on farmland spatiotemporal calibration units, thereby improving the accuracy and traceability of multi-source data quality analysis.

[0026] In some embodiments, multi-source farmland raw data is mapped to a farmland spatiotemporal calibration unit, quality features are extracted, and a data source quality profile is generated, including: Based on the collected metadata and observation objects carried by the original farmland data from multiple sources, the observation values ​​from each data source are matched to the corresponding farmland spatiotemporal calibration unit; The spatiotemporal benchmarks of the matched data sources are unified and the agronomic semantics are aligned to generate standardized farmland observation data. Based on standardized farmland observation data, quality characteristics that characterize collection stability, observation consistency, spatiotemporal representativeness, and equipment operating status are extracted. Based on the quality characteristics, the quality reliability and quality deviation characterization of each data source within the corresponding farmland spatiotemporal calibration unit are determined, and a data source quality profile associated with the data source identifier, the observed object, and the farmland spatiotemporal calibration unit is generated.

[0027] Specifically, after the target farmland has generated spatiotemporal calibration units, the system performs unit-level mapping on the accessed multi-source farmland raw data. For data reported by ground soil moisture acquisition terminals, the system reads the terminal number, installation coordinates, acquisition time, and observation object from the acquisition metadata, matches the observed values ​​such as soil moisture content and soil temperature to the spatial calibration object where the installation coordinates fall, and assigns them to the corresponding time calibration window according to the acquisition time. For UAV image inversion data, the system maps the observed values ​​such as vegetation index and canopy temperature to multiple farmland spatiotemporal calibration units according to the pixel coverage ratio, based on the coverage area after image orthorectification, imaging time, and inversion object.

[0028] For agricultural machinery operation data, the system determines the scope of impact of the operation based on trajectory points, operation width, and operation time, and maps observations such as fertilizer application, seeding density, and irrigation execution amount to the spatiotemporal calibration units of the farmland covered by the trajectory. For manual field inspection records, the system completes the mapping based on the sampling point location, entry time, and recording object, and retains the sampling radius and entry lag information.

[0029] Furthermore, after mapping, the system unifies the spatiotemporal references of the matched observations from various data sources. For spatial reference unification, the system converts sensor coordinates, image pixel coordinates, agricultural machinery trajectory coordinates, and manual sampling coordinates to the same farmland coordinate reference, and determines the spatial affiliation and representative range of the observations based on the spatial calibration object boundaries. For temporal reference unification, the system converts the acquisition times of different data sources to a unified time axis, performs window aggregation on high-frequency sampling data according to the time calibration window, marks effective time periods for low-frequency remote sensing data, and extends the impact time period for event-based agricultural machinery data. For example, irrigation operation data from 9:00 AM to 10:00 AM can be extended to the post-irrigation moisture response window, ensuring that subsequent soil moisture changes are consistent with the time sequence of management events.

[0030] During agronomic semantic alignment, the system maps observations from different data sources with different field names but reflecting the same farmland condition to a unified agronomic semantic space. For example, soil volumetric moisture content collected by ground sensors, water stress index obtained from remote sensing inversion, and water output recorded by irrigation equipment are all associated with farmland water status; UAV vegetation index, artificial seedling condition level, and crop canopy image recognition results are all associated with crop growth status. The system also performs semantic verification on observations based on crop type, growth stage, and management events to avoid misinterpreting normally changing growth indicators at different growth stages as data anomalies. After spatiotemporal benchmark unification and agronomic semantic alignment, the system generates standardized farmland observation data and retains the correspondence between the source, mapping unit, observation object, standardized value, and original value for each observation data.

[0031] Subsequently, the system extracts quality features based on standardized farmland observation data. For acquisition stability, the system calculates sampling interval fluctuations, duration of continuous missing measurements, magnitude of numerical jumps, and historical drift trends. For observation consistency, the system calculates numerical differences, trend differences, and response lags between different data sources within the same farmland spatiotemporal calibration unit. For spatiotemporal representativeness, the system combines sensor coverage radius, image pixel mixing ratio, trajectory coverage, and spatial distribution of sampling points to determine the representativeness of the observed values ​​for the farmland spatiotemporal calibration unit. For equipment operating status, the system reads status information such as battery power, communication quality, positioning accuracy, image cloud coverage ratio, and equipment calibration records to form equipment-side quality features. For example, if a soil moisture sensor samples normally for 6 consecutive hours but remains consistently higher than neighboring sensors, and there is no supporting rainfall or irrigation events, the system extracts a continuous offset feature; if a UAV image has localized shadow occlusion, the system extracts an insufficient effective image coverage feature.

[0032] Based on the aforementioned quality characteristics, the system calculates the quality reliability of each data source within its corresponding farmland spatiotemporal calibration unit and generates a quality deviation characterization. Quality reliability indicates the degree to which current observations can serve as a basis for subsequent quality judgments, while the quality deviation characterization describes the possible direction, intensity, and persistence of deviations. The system associates quality reliability and quality deviation characterization with the data source identifier, observation object, farmland spatiotemporal calibration unit, and acquisition time window to form a data source quality profile. For example, if a soil moisture acquisition terminal exhibits high acquisition stability, moderate observation consistency, and slight positive drift across three spatial calibration objects during the rice tillering stage, the data source quality profile records the terminal's reliability level and drift tendency within that growth stage and spatial range, providing a basis for subsequent selection of reliable anchor points and generation of self-calibration parameters.

[0033] Through the above processing, multi-source farmland raw data can be uniformly mapped to farmland spatiotemporal calibration units with spatial boundaries, time windows, and agronomic semantics. Based on standardized observations, a data source quality profile is formed, enabling the system to not only obtain farmland status data, but also to identify the credibility and deviation characteristics of each data source in a specific farmland scenario, thereby improving the accuracy of subsequent quality defect identification and deviation attribution.

[0034] In some embodiments, based on farmland spatiotemporal calibration units, agronomic process constraints, and data source quality profiles, a farmland quality constraint map characterizing observational correlations, state evolution, and quality transmission relationships is constructed, including: Using farmland spatiotemporal calibration units as map indexes, map nodes representing farmland status, data source observations, and data quality status are generated; Agronomic process constraints are converted into state evolution rules connecting different graph nodes, and the constraint relationships of farmland state changes with management and environmental processes are determined based on the state evolution rules. Configure quality attributes for the map nodes corresponding to each data source observation based on the data source quality profile, and establish the observation correlation between data source observation and farmland status; Based on observational correlations and constraint relationships, quality transmission parameters are determined, and farmland quality constraint maps are generated to support the selection of credible anchor points, residual analysis, and bias attribution.

[0035] Specifically, after generating farmland spatiotemporal calibration units and data source quality profiles, the system establishes a map index for farmland quality constraint maps based on these units. For each farmland spatiotemporal calibration unit, the system generates a corresponding unit index node and establishes farmland status nodes, data source observation nodes, and data quality status nodes under that unit index node. The farmland status node characterizes farmland conditions such as soil moisture, crop growth, canopy temperature, and pest and disease risk within that spatial range and time window. The data source observation node carries standardized observation data from data sources such as ground sensors, UAV imagery, agricultural machinery operation terminals, and manual field inspection records. The data quality status node records the quality reliability, deviation direction, deviation intensity, and deviation persistence in the data source quality profile. Taking a paddy field irrigation management area as an example, the same farmland spatiotemporal calibration unit can simultaneously include soil moisture sensor observation nodes, UAV canopy temperature observation nodes, irrigation execution observation nodes, and soil moisture status nodes, and these are uniformly organized through the unit index node.

[0036] When constructing state evolution relationships, the system reads agronomic process constraints and converts them into state evolution rules connecting different graph nodes. Agronomic process constraints can originate from crop growth models, agricultural management records, historical crop condition samples, and expert rule bases. The system extracts triggering conditions, response objects, response directions, response delays, and durations from the agronomic process constraints and encodes this information into state evolution rules in the graph.

[0037] For example, after irrigation is completed in the target farmland between 9 and 10 a.m., soil moisture typically rises shortly after irrigation, while canopy temperature may decrease as water stress decreases. Based on this, the system establishes state evolution edges with directional and temporal constraints between irrigation event nodes, soil moisture state nodes, and canopy temperature state nodes. Similarly, continuous high temperatures and low rainfall can intensify crop water stress; the system establishes delayed response relationships between environmental state nodes and crop growth state nodes accordingly. In this way, changes in farmland state are no longer simply viewed as numerical fluctuations, but rather organized as verifiable agronomic process responses.

[0038] Furthermore, the system configures quality attributes for data source observation nodes based on the data source quality profile. These quality attributes include current quality reliability, observation representativeness, equipment operational reliability, and historical deviation trends. For ground sensor observation nodes with stable sampling and consistency with neighboring observations, the system configures a higher quality reliability; for UAV image observation nodes obscured by cloud shadows, the system reduces the quality attribute corresponding to effective image coverage; and for agricultural machinery operation observation nodes with significant trajectory drift, the system reduces spatial representativeness.

[0039] Furthermore, after configuration, the system establishes observation associations based on the correspondence between observed objects and farmland conditions. For example, soil moisture sensor observation nodes are associated with soil moisture status nodes, UAV vegetation index observation nodes are associated with crop growth status nodes, and irrigation execution record observation nodes are associated with water management event nodes. The observation associations record the observation direction, observation weight, and applicable scope, enabling subsequent selection of reliable anchor points to identify which observation nodes are more suitable as references for the corresponding farmland conditions.

[0040] When determining quality transmission parameters, the system analyzes the impact path of quality defects in the map based on observation correlation and state evolution relationships. If a soil moisture sensor exhibits continuous positive drift, this defect will affect the judgment of soil moisture status and further impact irrigation demand assessment. Accordingly, the system configures quality transmission parameters between sensor observation nodes, soil moisture status nodes, and irrigation decision-related nodes. Quality transmission parameters can be jointly determined by observation weights, state response intensity, temporal distance, and spatial proximity, representing the degree of impact of a node's quality anomaly on the reliability of associated nodes. For localized low vegetation indices caused by cloud shadows in remote sensing images, the system only propagates the quality impact within the relevant spatiotemporal calibration units of farmland within the image coverage area, avoiding the spread of local image defects to unrelated areas.

[0041] After configuring the nodes, edges, and parameters as described above, the system generates a farmland quality constraint map. This map is organized around farmland spatiotemporal calibration units and simultaneously expresses the observational correlation between data source observations and farmland status, the state evolution relationship of farmland status changes with management and environmental processes, and the quality transmission relationship of data quality anomalies propagating along the observation chain. Subsequently, when selecting dynamic reliable anchor points, the system can prioritize observation nodes with high quality attributes that conform to the state evolution relationship; during residual analysis, it can determine whether data fluctuations are consistent with agronomic processes based on state evolution rules; and when attributing biases, it can trace back the source of anomalies along the quality transmission relationship.

[0042] Through the above processing, the system can unify scattered farmland observation data, agronomic process knowledge, and data source quality profiles into a computable farmland quality constraint map, transforming data quality judgment from single-point threshold judgment to correlation analysis under map constraints, thereby improving the rationality of credible anchor point selection, the accuracy of residual analysis, and the reliability of bias source attribution.

[0043] In some embodiments, agronomic process constraints are converted into state evolution rules connecting different graph nodes, and the constraint relationships of farmland state changes with management and environmental processes are determined according to the state evolution rules, including: Analyze the triggering conditions, response directions, and timing of action in agronomic process constraints that characterize changes in farmland state; The triggering conditions, response directions, and action timing are encoded as state evolution rules between graph nodes; According to the state evolution rules, state evolution edges with directional and temporal constraints are established between management event nodes, environmental state nodes, and farmland state nodes; The constraints on farmland state changes with management and environmental processes are determined based on state evolution edges.

[0044] Specifically, in constructing the farmland quality constraint map, the system reads agronomic process constraints from the agronomic management platform, crop growth model, historical crop condition samples, and expert rule base, and performs structured analysis on the constraints related to changes in farmland status. The system first identifies triggering conditions, which can include management event triggering conditions and environmental process triggering conditions.

[0045] In some examples, management event triggers are formed from agricultural records such as irrigation start, fertilization completion, plant protection operations, and agricultural machinery compaction; environmental process triggers are formed from field meteorological events such as rainfall, high temperature, low temperature, continuous cloudy days, and strong winds. The system further analyzes the response direction and timing of each trigger condition. The response direction is used to characterize the increase, decrease, stabilization, or enhanced fluctuation of farmland status after the trigger condition occurs, while the timing of action is used to characterize the response start time, response peak time, and response duration.

[0046] Taking paddy field water management as an example, if irrigation equipment records show that irrigation was completed in a certain irrigation management area between 9:00 AM and 10:00 AM, the system will interpret the completion of irrigation as the trigger condition, and the increase in soil moisture, decrease in canopy temperature, or reduction in water stress as the response direction. The system will also define 0-2 hours after irrigation as the rapid response phase and 2-8 hours as the diffusion stabilization phase. If the field weather station records continuous high temperatures and low rainfall, the system will interpret this as the environmental process trigger condition, and the decrease in soil moisture, increase in canopy temperature, and decrease in the rate of increase in crop growth indicators as the response direction. The system will configure the timing of these actions as a daily cumulative response. For fertilization events, the system can interpret a short-term increase in soil electrical conductivity and a delayed improvement in crop growth indicators as corresponding response relationships. Through the above analysis, the system can convert agronomic knowledge, originally existing in the form of natural language or management records, into computable process constraints.

[0047] After parsing, the system encodes the triggering conditions, response direction, and timing of action into state evolution rules between graph nodes. Each state evolution rule includes at least the triggering node type, target state node type, response direction parameter, time delay parameter, duration parameter, and applicable agronomic state identifier. The system adapts and filters the state evolution rules based on the agronomic state identifiers configured in the farmland spatiotemporal calibration unit. For example, even with the same rainfall event, the soil moisture response intensity differs between the seedling stage of dryland maize and the tillering stage of rice. The system adjusts the response direction parameter and duration parameter based on crop type, growth stage, and soil texture to ensure the state evolution rules conform to the agronomic scenario of the current spatial calibration object.

[0048] Subsequently, the system establishes state evolution edges in the farmland quality constraint map. For state changes caused by management events, the system establishes directed state evolution edges between management event nodes and farmland state nodes; for state changes caused by meteorological conditions, the system establishes directed state evolution edges between environmental state nodes and farmland state nodes; for situations where one farmland state affects another, the system establishes state linkage edges between relevant farmland state nodes. Each state evolution edge carries directional and temporal constraints. The directional constraint limits the reasonable direction of change of the target farmland state after the triggering condition occurs, and the temporal constraint limits the effective time range of the change. If the irrigation event node points to the soil moisture state node, the directional constraint is moisture increase, and the temporal constraint is several hours after irrigation ends; if the high temperature event node points to the canopy temperature state node, the directional constraint is temperature increase, and the temporal constraint is the duration of high temperature and the period of lag effect.

[0049] In subsequent data quality analysis, the system utilizes state evolution edges to determine the constraints on farmland status changes with management and environmental processes. When the soil moisture sensor reading in a farmland spatiotemporal calibration unit rises after irrigation, and the rise time falls within the temporal constraint range of the corresponding state evolution edge, the system identifies this change as a reasonable response consistent with the agronomic process. When the soil moisture sensor reading suddenly increases under conditions of no rainfall, no irrigation, and no synchronous changes in the adjacent area, and this change does not meet the triggering conditions and timing of any state evolution edge, the system treats this change as a potential quality anomaly and includes it in subsequent residual analysis. For remote sensing growth indicators, if there is no immediate increase after fertilization but a gradual improvement within the subsequent growth window, the system judges it as a reasonable delayed response based on the temporal constraint between the fertilization event and the crop growth status node, avoiding misjudgment as a remote sensing inversion anomaly.

[0050] Through the above methods, the system transforms agronomic process constraints into state evolution rules with directional, temporal, and scenario-adaptive capabilities, forming computable constraint relationships between management event nodes, environmental state nodes, and farmland state nodes. This processing enables the farmland quality constraint map to possess agronomic process understanding capabilities, distinguishing between real farmland state changes and data quality defects when data fluctuations occur, thereby improving the accuracy of subsequent reliable anchor point selection, process constraint residual calculation, and deviation source attribution.

[0051] In some embodiments, a dynamic reliable anchor point is determined based on the data source quality profile and farmland quality constraint map, and a corresponding self-calibration reference state is generated, including: The quality reliability of each data source within the corresponding farmland spatiotemporal calibration unit is determined based on the data source quality profile. Based on the farmland quality constraint map, calculate the anchor point fit of each data source observation relative to the farmland state evolution relationship and observation correlation relationship; Data source observations that meet the anchor point selection criteria are identified as dynamic trusted anchor points, and the dynamic trusted anchor points are fused based on the anchor point fit. When there are insufficient dynamic reliable anchor points, state inference is performed based on the state evolution relationship in the farmland quality constraint map to generate the self-calibration reference state of the corresponding farmland spatiotemporal calibration unit.

[0052] Specifically, after the farmland quality constraint map of the target farmland has been generated, the system uses the farmland spatiotemporal calibration unit as the processing granularity to perform reliable anchor point screening on the observations of each data source that enters the same spatial range and temporal calibration window. The system first reads the quality reliability, acquisition stability, observation representativeness, equipment operation reliability, and historical deviation trend recorded in the data source quality profile, and combines this with the agronomic status identifier of the current farmland spatiotemporal calibration unit to determine the usability level of each data source in the current scenario.

[0053] For example, in an irrigation management area during the rice tillering stage, if a soil moisture sensor reports continuously and normally, has stable communication quality, and has a small long-term deviation from adjacent sensors, the system will configure the sensor with a higher quality reliability under the soil moisture observation target. If the area corresponding to a UAV image has shadow occlusion or a high pixel mixing ratio, the quality reliability of the image will be reduced under the crop growth observation target.

[0054] After the quality reliability is determined, the system calculates the anchor point fit of observations from each data source based on the farmland quality constraint map. The system first locates the observation nodes of each data source in the map and finds the farmland state nodes associated with the observation nodes. Then, it verifies whether the observations conform to the current agronomic process along the state evolution relationship. For example, if the irrigation event node is triggered at 9:00 AM, the state evolution edge indicates that the soil moisture state should rise within 0 to 2 hours after irrigation. In this case, the soil moisture observations that rise synchronously after irrigation have a high degree of evolution consistency. If a sensor shows a sudden increase in soil moisture under conditions of no rainfall, no irrigation, and stable neighboring areas, the evolution consistency decreases.

[0055] The system also compares the observation trends of the same farmland state from different data sources within the same spatiotemporal calibration unit along the observation correlation. For example, if there is a consistent response between soil moisture increase, canopy temperature decrease, and water stress index decrease, the corresponding observation values ​​have a high degree of observation consistency. The system weights and fuses quality reliability, evolution consistency, and observation consistency to obtain the anchor point fit.

[0056] The system then determines dynamic reliable anchor points based on anchor point selection criteria. These criteria may include a quality reliability reaching a preset reliability level, an anchor point fit exceeding an anchor point threshold, the observation not being in a persistently abnormal state, and the observation node covering the current farmland state node. Observations meeting these criteria are marked as dynamic reliable anchor points. In cases where multiple dynamic reliable anchor points exist within the same farmland spatiotemporal calibration unit, the system performs weighted fusion based on anchor point fit to generate a self-calibration reference state.

[0057] For example, in a rice irrigation management area, the ground soil moisture sensor reading is 28%, the interpolated reference value for adjacent areas is 27.5%, and the UAV water stress inversion result corresponds to a low stress state. The system integrates these values ​​based on the adaptability of their respective anchor points to generate a soil moisture reference state within that time calibration window. For crop growth status, the system can integrate UAV vegetation index, manual field inspection crop condition levels, and historical growth evolution results to generate a growth reference state.

[0058] When there are insufficient dynamic reliable anchor points, the system does not directly use low-reliability data as a reference. Instead, it infers the state based on the state evolution relationship in the farmland quality constraint map. Starting from the reliable farmland state within the previous time calibration window, the system performs state propagation by combining environmental state nodes, management event nodes, and spatial adjacency nodes to obtain the inferred state of the current farmland spatiotemporal calibration unit.

[0059] For example, if a plot of land lacks effective drone imagery due to rainy weather and some sensor communications are interrupted, the system can infer the current soil moisture reference state based on the previous day's soil moisture status, the current day's rainfall, evapotranspiration estimates, the moisture status of adjacent plots, and irrigation records. For crop growth, the system can infer the current growth reference state based on the growth stage, accumulated temperature changes, and historical growth curves, and assign an inference confidence level to this reference state to avoid equating the inference results with high-confidence anchor point results.

[0060] After generating the self-calibration reference state, the system associates and saves it with the farmland spatiotemporal calibration unit, the observed object, the time calibration window, the participating anchor points, and the anchor point fit, providing a reference basis for subsequent cross-source consistency residual calculation and process constraint residual calculation. In this way, the system does not rely on fixed truth values ​​or a single reference data source, but dynamically determines a reliable reference based on the data source quality profile and farmland quality constraint map. This allows the system to maintain the stability of the reference state even under conditions of sensor drift, remote sensing occlusion, or local data loss, thereby improving the accuracy of data quality defect identification and the reliability of farmland multi-source data self-calibration.

[0061] In some embodiments, based on the farmland quality constraint map, the anchor point fit of each data source observation relative to the farmland state evolution relationship and observation correlation relationship is calculated, including: In the farmland quality constraint map, determine the observation nodes corresponding to the observation values ​​of each data source and the associated farmland status nodes; Based on the state evolution relationship of farmland state nodes, the evolution consistency between the observation values ​​of each data source and the current farmland state change trend is calculated; Based on the observation correlation relationship of the observation node connection, the observation consistency between the observation values ​​of each data source and the related observation values ​​within the same farmland spatiotemporal calibration unit is calculated. Based on evolutionary consistency, observational consistency, and the quality reliability of the corresponding data source, the anchor point fit of observations from each data source is generated.

[0062] Specifically, in a farmland quality constraint map of a wheat-growing area, after receiving soil moisture sensor readings, field weather station data, UAV vegetation index inversion results, and manual field inspection records from the same farmland spatiotemporal calibration unit, the system first locates the corresponding observation nodes in the farmland quality constraint map based on the data source identifier, observation object, and unit identifier in the standardized farmland observation data. For example, the soil moisture sensor observation value numbered S01 corresponds to the soil moisture observation node, the water stress index obtained from UAV image inversion corresponds to the crop water status observation node, and the rainfall and evapotranspiration estimates recorded by the weather station correspond to the environmental status observation node. The system further determines the farmland status nodes associated with each observation node based on the mapping relationship of the observation objects, such as soil moisture status nodes, crop growth status nodes, and water stress status nodes, and reads the state evolution edges and observation association edges corresponding to each node within the current time calibration window.

[0063] When calculating evolutionary consistency, the system uses the state evolution relationship connecting farmland state nodes as a constraint to determine whether the observed values ​​from each data source conform to the current farmland state change trend. The state evolution relationship can be jointly determined by management events, environmental events, and historical states. For example, if an irrigation event occurs in the target unit between 8:00 AM and 9:00 AM, the state evolution edge from the irrigation event node to the soil moisture state node in the graph indicates that soil moisture should show an upward trend within 2 hours after irrigation, while the state evolution edge from the irrigation event node to the canopy temperature state node indicates that canopy temperature may show a downward trend. If the soil moisture reported by the S01 soil moisture sensor at 9:30 AM is higher than the previous window, and the increase is within the irrigation response range, the system determines that the observed value is consistent with the farmland state change trend and assigns a higher evolutionary consistency. If another sensor shows a sudden increase in soil moisture under conditions of no irrigation, no rainfall, and strong evapotranspiration, the system determines that the observed value is inconsistent with the state evolution relationship and reduces the corresponding evolutionary consistency.

[0064] When calculating observational consistency, the system uses the observational correlation relationships between observation nodes as a constraint, comparing the consistency of observations reflecting the same or related farmland conditions within the same farmland spatiotemporal calibration unit. For soil moisture status, the system compares the response relationships between soil moisture sensor readings, irrigation execution amounts, moisture changes in adjacent spatial calibration objects, and UAV water stress inversion results. If the soil moisture sensor shows an increase in moisture, and the irrigation execution record is valid, a similar increase is observed in neighboring areas, and the UAV water stress index decreases, then the observation value has a high degree of observational consistency with the associated observation values.

[0065] In some examples, if an observation deviates independently while other observation nodes within the same time window do not show corresponding changes, the system reduces the observation consistency of that observation based on the weight of the observation association edges and the relevance of the observed objects. For crop growth status, the system can compare the trend consistency between UAV vegetation index, artificial seedling condition level, and historical growth curves, thereby avoiding relying solely on a single image result as a reliable reference.

[0066] Furthermore, the system reads the quality reliability of the corresponding data source from the data source quality profile and merges it with the evolutionary consistency and observational consistency to generate an anchor point fit. Quality reliability reflects the data source's acquisition stability, observational representativeness, and equipment operating status within the current farmland spatiotemporal calibration unit; evolutionary consistency reflects whether the observed values ​​conform to the state change trends derived from the agronomic process; and observational consistency reflects whether the observed values ​​are supported by other related observations. The system can set different fusion weights according to the observed object. For example, for soil moisture status, the weights of ground sensor quality reliability and irrigation response evolutionary consistency are increased; for crop growth status, the weights of remote sensing image observational consistency and growth period evolutionary consistency are increased. If the S01 soil moisture sensor quality reliability is high, and both evolutionary consistency and observational consistency meet the anchor point fit conditions, the system generates a high anchor point fit and uses this observed value as a candidate for a dynamic reliable anchor point.

[0067] When observations from multiple data sources are used simultaneously in anchor point fit calculations, the system also incorporates spatial adjacency and temporal continuity relationships from the farmland quality constraint map to perform a smoothing check on the anchor point fit. For example, adjacent spatial calibration objects within the same irrigation zone may have similar water responses. If an observation, while currently showing high consistency, deviates significantly from the trend of changes in the preceding and following time windows, the system reduces the final anchor point fit for that observation. If a UAV image observation is affected by local shading, even if it is close to some manual records, the system will reduce the anchor point fit based on image quality attributes. Therefore, anchor point fit is not simply determined by numerical similarity, but rather by a comprehensive consideration of the rationality of the agronomic process, cross-source observation support, and the reliability of the data source itself.

[0068] Through the above processing, the system can accurately identify the observation nodes and farmland state nodes corresponding to the observation values ​​of each data source in the farmland quality constraint map, and calculate the anchor point fit degree by using the state evolution relationship and observation correlation relationship, so that the determination of dynamic reliable anchor points has the basis of agronomic constraints and cross-source consistency, thereby improving the reliability of self-calibration reference state and reducing the risk of abnormal data being mistakenly selected as reliable reference.

[0069] In some embodiments, data quality defect types are identified based on the cross-source consistency residuals between observations from each data source and the self-calibration reference state, and the process constraint residuals between observations from each data source and agronomic process constraints, including: The observations from each data source are compared with the self-calibration reference state of the corresponding farmland spatiotemporal calibration unit to generate cross-source consistency residuals. Based on agronomic process constraints, the state changes of observations from each data source within the corresponding time calibration window are constrained and verified to generate process constraint residuals. By integrating cross-source consistency residuals, process constraint residuals, and data source quality profiles, a quality anomaly discrimination characterization is generated. Input the quality anomaly discrimination characterization into the quality defect identification model to determine the types of data quality defects associated with the data source, the observed object, and the farmland spatiotemporal calibration unit.

[0070] Specifically, after the spatiotemporal calibration unit of the target farmland has established a self-calibration reference state, the system performs residual analysis on the observations from each data source that enter the same time calibration window. Taking a rice irrigation management area as an example, the self-calibration reference state of this unit between 10:00 AM and 11:00 AM indicates that soil moisture should be around 28%, canopy temperature should be lower than before irrigation, and water stress index should be reduced. The system reads the soil moisture sensor observations, UAV thermal infrared inversion values, irrigation execution records, and field meteorological data within this time calibration window, and compares the observations from each data source with the self-calibration reference state under the corresponding observation object.

[0071] For numerical observations, the system calculates the amplitude difference, direction of change difference, and time response difference between the observations and the reference state; for image inversion observations, the system calculates the spatial texture difference and coverage area difference between the inversion results and the reference state; for event record observations, the system calculates the matching difference between event execution amount, execution time, and state response, thereby generating cross-source consistency residuals.

[0072] After generating cross-source consistency residuals, the system further calculates process constraint residuals based on agronomic process constraints. The system reads the state evolution edges associated with the current farmland spatiotemporal calibration unit in the farmland quality constraint map, determines whether there are triggering conditions such as irrigation, rainfall, fertilization, high temperature, and pests and diseases in the current time calibration window, and judges whether the state changes of the observations from each data source after the occurrence of the triggering conditions satisfy the response direction and action sequence.

[0073] For example, soil moisture should increase reasonably and canopy temperature should decrease reasonably within 2 hours after irrigation. If the soil moisture sensor does not show a significant increase after irrigation, but the irrigation execution record and adjacent units all show an effective moisture response, then the observed soil moisture value will have a large process constraint residual. If a UAV image shows a low vegetation index after consecutive cloudy days, but the image quality characteristics show a high proportion of cloud shadows, and manual field inspection does not find a decline in crop growth, then the observed image value will have a large process constraint residual regarding crop growth status.

[0074] The system fuses cross-source consistency residuals, process constraint residuals, and data source quality profiles to generate a quality anomaly discrimination characterization. During fusion, the system records not only the residual magnitude but also the residual direction, residual duration, residual location, and the quality status of the corresponding observation object and data source. For example, if a soil moisture sensor consistently shows a higher value relative to its self-calibration reference state across multiple consecutive time calibration windows, with the magnitude of the deviation gradually increasing, while the device communication is normal and the sampling interval is stable, the system generates a quality anomaly discrimination characterization representing a continuously positive shift.

[0075] In some examples, if a sensor continuously outputs the same value and does not change with rainfall, irrigation, or evapotranspiration, the system generates a quality anomaly discrimination characterization characterizing the abnormal card value. If there is a systematic offset between the operation trajectory of an agricultural machine and the boundary of the plot, and the corresponding operation influence range is inconsistent with the subsequent state response location, the system generates a quality anomaly discrimination characterization characterizing the spatial positioning deviation.

[0076] Subsequently, the system inputs the quality anomaly discrimination characterization into the quality defect identification model. The quality defect identification model can be implemented using a graph-constrained temporal classification model. The model input includes residual sequences, map neighborhood features, quality confidence, and agronomic status indicators. The model output includes data quality defect types and defect confidence levels. The model first performs temporal deviation analysis on the quality anomaly discrimination characterization, identifying time patterns such as sudden spikes, continuous drift, periodic missing data, and response lag. Then, based on the observation correlations and state evolution relationships in the farmland quality constraint map, it performs neighborhood verification to determine whether the anomaly occurs only in a single data source or propagates along the same equipment, the same image coverage area, or the same operational trajectory.

[0077] If the anomaly occurs only in a single soil moisture sensor and continues to deviate from adjacent units, the model outputs a sensor drift type; if the anomaly is concentrated in the image shadow area, the model outputs a remote sensing inversion contamination type; if the anomaly is consistent with the overall translation of the agricultural machinery trajectory, the model outputs a trajectory positioning offset type; if the anomaly is caused by the misalignment of the acquisition time and the agronomic response window, the model outputs a timestamp offset type.

[0078] The system associates and saves the identified data quality defect types with the data source identifier, observation object, farmland spatiotemporal calibration unit, and time calibration window, and writes the defect confidence level into the subsequent deviation attribution process. For example, the system can record that the soil moisture sensor numbered S01 has proportional drift under the soil moisture observation object in a certain rice irrigation management area, with a confidence level of 0.86; it can also record that a UAV flight has cloud shadow interference under the crop growth inversion object in a local plot, with a confidence level of 0.79. Through the above processing, the system can identify data quality defect types by comprehensively considering cross-source consistency, agronomic process rationality, and data source quality profile, avoiding misjudging real agronomic responses as anomalies, and improving the accuracy of subsequent deviation source attribution and data source generation from calibration parameters.

[0079] In some embodiments, the quality anomaly discrimination characterization is input into the quality defect identification model to determine the types of data quality defects associated with the data source, the observed object, and the farmland spatiotemporal calibration unit, including: A quality defect identification model is used to perform temporal deviation analysis and graph neighborhood analysis on the quality anomaly discrimination characterization to generate candidate defect characterizations. Based on the matching relationship between candidate defect representations and preset quality defect patterns, the corresponding data quality defect type is determined; Based on the residual strength in the quality anomaly discrimination characterization and the quality confidence in the data source quality profile, the defect confidence of the data quality defect type is determined. The data quality defect type and defect confidence level are associated with the corresponding data source, observation object, and farmland spatiotemporal calibration unit for output.

[0080] Specifically, the quality defect identification model can be deployed in farmland data quality analysis services. The model takes quality anomaly discrimination characterization as input and outputs data quality defect type and defect confidence level. The quality anomaly discrimination characterization includes cross-source consistency residuals, process constraint residuals, residual direction, residual duration, residual spatial range, observation object, data source quality confidence level, and agronomic status identifiers of farmland spatiotemporal calibration units. Before inputting the data into the model, the system first sequentially organizes the quality anomaly discrimination characterization according to the data source identifier and observation object, enabling continuous analysis of anomalous changes from the same data source within a continuous time calibration window. Simultaneously, the spectral neighborhood information of the corresponding observation nodes is also input into the model.

[0081] In the time-series deviation analysis phase, the quality defect identification model reads the residual sequence of the target data source within multiple time calibration windows and analyzes the residual change patterns. If the residual suddenly increases and recovers rapidly within a short period, the model identifies it as a peak candidate pattern; if the residual continuously shifts in the same direction and the shift amplitude gradually increases, the model identifies it as a drift candidate pattern; if the observed value remains unchanged for a long time during changes in conditions such as rainfall, irrigation, and evapotranspiration, the model identifies it as a card value candidate pattern; if the direction of change of the observed value is correct but the response time lags significantly relative to the agronomic process constraints, the model identifies it as a time-shift candidate pattern. For example, if a soil moisture sensor consistently exceeds the self-calibration reference state by 3% to 5% for 5 consecutive hours, and the deviation gradually increases, the model generates defect candidate representations corresponding to the positive drift of the sensor in the time-series deviation analysis.

[0082] During the neighborhood analysis phase, the model reads the farmland state nodes, management event nodes, environmental state nodes, and adjacent observation nodes connected to the target observation node based on the farmland quality constraint graph, and determines whether the anomaly can be explained by other nodes in the neighborhood. If a soil moisture observation value is high, and adjacent sensors, irrigation execution records, and UAV water stress inversion results within the same irrigation zone all support an increase in moisture, the model reduces the likelihood of this anomaly as a quality defect. If only a single sensor shifts, and adjacent nodes and state evolution edges do not support this change, the model strengthens the defect candidate representation of single-point equipment anomalies.

[0083] For remote sensing image data, if the abnormal area highly overlaps with the cloud shadow quality node or the image edge area, the model will point to the remote sensing inversion contamination as a candidate defect. For agricultural machinery operation data, if there is a systematic misalignment between the trajectory observation node and the plot boundary, the operation width, and the subsequent state response location, the model will point to the trajectory positioning offset as a candidate defect.

[0084] Subsequently, the system matches candidate defect representations with preset quality defect patterns. These preset quality defect patterns can be constructed from historical anomaly samples, equipment calibration records, manual review results, and agronomic expert rules, and are saved as pattern templates. Each pattern template includes residual morphology, persistent features, spectral neighborhood response, applicable data source type, and applicable observation object. The model calculates the degree of matching between candidate defect representations and each pattern template, and selects the data quality defect type corresponding to the pattern whose matching degree meets the conditions.

[0085] For example, candidate representations that show a continuous positive offset and normal equipment operation match sensor proportional drift type; candidate representations that show continuous repetitive output and whose farmland status should change match acquisition card value type; candidate representations that show the overall observation falling into the wrong time calibration window and swapping responses with the previous and next windows match timestamp offset type; and candidate representations that show image inversion anomalies consistent with low effective coverage areas match image quality interference type.

[0086] Furthermore, after determining the type of data quality defect, the system calculates the defect confidence level based on the residual intensity in the quality anomaly discrimination characterization and the quality reliability in the data source quality profile. The higher the residual intensity, the longer its duration, and the more significant the conflict with agronomic process constraints, the higher the defect confidence level. If the historical quality reliability of the data source is low, and the current equipment operation status exhibits communication anomalies, positioning anomalies, or image quality anomalies, the defect confidence level is further increased. If the historical quality reliability of the data source is high, and there are management events or environmental events in the map neighborhood that can explain the residual, the system lowers the defect confidence level to avoid misjudging real farmland state changes as data quality defects. For example, although an increase in water content after irrigation produces a large numerical difference, it conforms to the state evolution rules, and the system will not directly determine it as a sensor anomaly; while a continuous offset without triggering events will obtain a higher defect confidence level.

[0087] Finally, the system correlates the data quality defect type, defect confidence level, defect candidate characteristics, and discrimination criteria with the corresponding data source, observation object, and farmland spatiotemporal calibration unit to output the quality defect identification result. This result can be recorded as: the soil moisture sensor (S01) exhibits sensor drift under the soil moisture observation object in the target irrigation management area, with a defect confidence level of 0.88; or it can be recorded as: a UAV flight exhibits image cloud shadow interference under the wheat growth observation object, with a defect confidence level of 0.81. Through the above processing, the system can identify the data quality defect type while providing the confidence level and associated objects, providing a clear basis for subsequent bias attribution and data source generation from calibration parameters.

[0088] In some embodiments, the source of deviation is attributed to the data quality defect type based on the farmland quality constraint map, and deviation attribution results are generated, including: Based on the type of data quality defect, identify the corresponding abnormal observation nodes and associated quality status nodes in the farmland quality constraint map; Starting from the abnormal observation nodes, a path search is performed along the observation correlations, state evolutions, and quality transmission relationships in the farmland quality constraint map to generate candidate deviation source paths; The corresponding source contribution is determined based on the residual explanation capability and quality transmission parameters of each candidate bias source path; The target deviation source path is determined from the candidate deviation source paths based on the source contribution, and deviation attribution results containing deviation source nodes, attribution paths, and contribution characterizations are generated.

[0089] Specifically, after the quality defect identification model outputs the data quality defect type, the system uses the farmland quality constraint map as the attribution carrier to conduct a retrospective analysis of the defect location and possible source. Taking a rice irrigation management area as an example, the soil moisture sensor numbered S01 was consistently higher than the self-calibration reference state for six consecutive time calibration windows, and the quality defect identification model determined it to be sensor proportional drift.

[0090] Based on the data quality defect type, the system locates the soil moisture observation node corresponding to S01 in the farmland quality constraint map and marks this observation node as an anomalous observation node. Simultaneously, the system reads the data quality status nodes connected to this observation node, identifying states such as decreased quality confidence, increased positive bias, and prolonged bias duration as associated quality status nodes. If the anomaly originates from UAV image inversion results, the system locates the corresponding image observation node and image quality status node, recording quality states such as cloud cover, pixel mixing, and decreased inversion confidence.

[0091] Furthermore, after identifying the anomalous observation nodes, the system uses these nodes as the starting point for backtracking and performs path searches along different relational edges in the farmland quality constraint map. Along the observation correlations, the system searches for farmland state nodes associated with the anomalous observation nodes, as well as other observation nodes within the same farmland spatiotemporal calibration unit reflecting the same farmland state.

[0092] For example, the S01 soil moisture observation node is associated with the soil moisture status node, and also has observational associations with adjacent soil moisture sensor observation nodes, irrigation execution observation nodes, and UAV water stress observation nodes. Along the state evolution relationship, the system searches for management event nodes or environmental status nodes within the current time window that can explain soil moisture changes, such as rainfall nodes, irrigation event nodes, and evapotranspiration enhancement nodes. Along the mass transmission relationship, the system continues to track whether anomalies may be transmitted from equipment operating status, communication gateway status, spatial positioning status, or upstream data processing status.

[0093] The system generates multiple candidate deviation source paths based on the path search results. For the example of a persistently high S01 sensor reading, candidate deviation source paths may include a device drift path tracing back from the S01 observation node to the S01 device quality status node, a management event interpretation path tracing back from the S01 observation node to the irrigation event node, a communication anomaly path tracing back from the S01 observation node to the same gateway communication status node, and a spatial local disturbance path tracing back from the S01 observation node to an adjacent spatial calibration object. The system does not directly determine all candidate paths as attribution results, but instead calculates the residual interpretation capability for each candidate deviation source path.

[0094] Residual interpretability indicates whether a path can explain the direction, strength, persistence, and spatial range of residuals formed by anomalous observations relative to the self-calibration reference state. If an irrigation event can explain the increase in soil moisture but cannot explain the persistent positive deviation of S01 being higher than that of adjacent sensors, then the residual interpretability of the irrigation event path is low. If there is a historical drift trend in the S01 equipment quality status, and the deviation direction is consistent with the current residual direction, then the equipment drift path has high residual interpretability.

[0095] Furthermore, the system calculates the source contribution by combining the quality transmission parameters in the farmland quality constraint map. The quality transmission parameters reflect the strength of the quality impact of the deviation source node on the abnormal observation node, and can be determined by edge weights, temporal distance, spatial distance, observation correlation, and quality state strength. If there is a direct quality transmission edge between the abnormal observation node and the equipment quality state node, and the data source quality profile records long-term positive sensor drift, then the source contribution corresponding to the equipment quality state node is relatively high.

[0096] In some examples, if a communication gateway affects multiple sensors simultaneously, and all sensors exhibit synchronization failures or anomalies, the contribution of the communication gateway node increases; conversely, if only a single sensor is abnormal while other sensors under the same gateway are functioning normally, the contribution of the communication gateway path decreases. For remote sensing image anomalies, if the anomaly area highly overlaps with the coverage area of ​​cloud shadow quality nodes, the source contribution of cloud shadow quality nodes is higher than that of the actual crop growth change path.

[0097] The system sorts candidate deviation source paths based on their contribution and determines the target deviation source path according to preset attribution confirmation conditions. If the highest source contribution is significantly higher than other paths, the system determines the corresponding path as the primary attribution path; if the contributions of multiple paths are similar, the system can determine them as joint attribution paths and record their respective contribution percentages.

[0098] For example, in the case of an anomaly in the S01 soil moisture sensor, the contribution of the equipment drift path is 0.72, the contribution of the irrigation event interpretation path is 0.16, the contribution of the communication anomaly path is 0.08, and the contribution of the spatial local disturbance path is 0.04. The system determines the equipment drift path as the source of the target deviation and uses the S01 equipment quality status node as the deviation source node. For anomalies in UAV imagery, if the contribution of the cloud shadow quality node is 0.65, the contribution of the plot boundary registration error node is 0.22, and the contribution of the actual growth change path is 0.13, the system determines the cloud shadow interference as the main source of deviation, while retaining the registration error as a secondary source.

[0099] Finally, the system generates deviation attribution results and associates and saves the deviation source node, attribution path, source contribution, residual explanation capability, associated data quality defect type, and farmland spatiotemporal calibration unit. The attribution results can be used to generate subsequent data source calibration parameters, such as generating proportional correction parameters for sensor drift, effective coverage correction parameters for image cloud shadow interference, and spatial translation correction parameters for agricultural machinery trajectory offset.

[0100] Through the above processing, the system can further trace the specific source of the deviation from the data quality defect type, transforming the anomaly handling from simply identifying defects to locating the source under the constraints of the map path, thereby improving the reliability of deviation attribution and the accuracy of subsequent calibration parameter generation.

[0101] In some embodiments, a data source is generated based on the deviation attribution result and calibration parameters are used to correct the original multi-source farmland data, generating a reliable farmland data view, including: Determine the mode of action and scope of influence of the target data source based on the deviation attribution results; Based on the mode of deviation action, the scope of deviation influence, and the type of data quality defects, data sources associated with data sources, observed objects, and farmland spatiotemporal calibration units are generated from calibration parameters. By using data sources to perform observation correction and confidence reassessment on raw multi-source farmland data using calibration parameters, calibrated farmland data is generated. The calibrated farmland data is correlated with the corresponding original observations, data sources, calibration parameters, bias attribution results, and confidence characterization to generate a reliable farmland data view.

[0102] Specifically, after completing the deviation source attribution, the system reads the deviation source node, target deviation source path, source contribution, and residual interpretation capability recorded in the deviation attribution results, and determines the target data source that needs to be calibrated. Taking the soil moisture sensor S01 in a rice irrigation management area as an example, the deviation attribution results show that the anomaly of the S01 soil moisture observation node mainly originates from the proportional drift in the equipment quality status node, with a source contribution of 0.72. The anomaly's influence range covers three farmland spatiotemporal calibration units within the effective coverage radius of the sensor, and the affected object is soil moisture status. Based on this, the system determines that the target data source is the S01 soil moisture sensor, the deviation mechanism is a proportional positive deviation that amplifies with the increase of the observed value, and the deviation's influence range includes the corresponding spatial calibration object, the soil moisture observation object, and the continuous anomaly time calibration window.

[0103] For different sources of deviation, the system employs different methods to determine the mode of deviation action and the scope of its influence. If the deviation attribution result points to sensor zero-point drift, the system determines the mode of deviation action as an additive offset in a fixed direction; if it points to sensor proportional drift, the system determines the mode of deviation action as a proportional offset related to the amplitude of the observed value; if it points to remote sensing cloud shadow interference, the system determines the mode of deviation action as local inversion downsampling or upsampling within the image coverage area; if it points to agricultural machinery trajectory positioning offset, the system determines the mode of deviation action as a systematic spatial translation; if it points to timestamp offset, the system determines the mode of deviation action as a misalignment between the observed value and the time calibration window. The system further limits the scope of deviation influence by combining the spatial boundaries, time windows, and observation correlations in the farmland quality constraint map, preventing the abnormal extension of local data sources to unrelated plots or unrelated observation objects.

[0104] Subsequently, the system generates data source calibration parameters based on the deviation action mode, deviation influence range, and data quality defect type. For the S01 proportional drift example, the system fits the proportional correction coefficient and intercept compensation amount according to the residual sequence between the abnormal observation value and the self-calibration reference state, and configures applicable conditions for the correction parameter, including the data source identifier S01, soil moisture observation object, corresponding farmland spatiotemporal calibration unit, abnormal time period, and rice tillering stage agronomic status identifier.

[0105] For remote sensing cloud shadow interference, the system generates image effective coverage correction parameters, cloud shadow area weight attenuation parameters, and inversion value compensation parameters; for agricultural machinery trajectory positioning offset, the system generates trajectory translation, operating width correction, and spatial remapping rules; for manual recording lag, the system generates time window reregistration parameters and sampling representativeness weights. Each set of data originates from calibration parameters and is associated with the deviation attribution results for easy tracing of parameter sources later.

[0106] When correcting raw data from multiple farmland sources, the system performs correction on the raw observations corresponding to the target data source according to the calibration parameters from which the data source originates. For the S01 soil moisture sensor, the system performs proportional compensation and intercept correction on the raw soil moisture readings that fall within the range affected by the deviation, and recalculates the reliability based on the residual change between the corrected observations and the self-calibration reference state.

[0107] For example, if the original reading is 33%, while the reasonable state supported by the self-calibration reference state and adjacent observations is approximately 29%, the system corrects the reading to 29.3% based on the proportional correction factor, and simultaneously adjusts the confidence level of the data from low confidence to medium-high confidence after calibration. For remote sensing inversion data, the system reduces the weight of contaminated pixels within the cloud shadow coverage area and compensates using adjacent effective pixels and historical growth trends; for agricultural machinery operation data, the system redetermines the operation coverage unit based on the trajectory translation correction parameters and updates the spatial assignment of the corresponding fertilizer or irrigation amount.

[0108] Furthermore, after completing the observation correction and confidence reassessment, the system generates calibrated farmland data and associates this data with the original observations, data sources, calibration parameters, bias attribution results, and confidence characterization to form a reliable farmland data view. The reliable farmland data view retains the original values, calibration values, data source identifiers, farmland spatiotemporal calibration units, observation objects, defect types, attribution paths, calibration parameter versions, and post-calibration confidence. For example, when the irrigation decision module reads soil moisture status, it no longer directly calls the original reading of S01, but instead calls the calibrated soil moisture data from the reliable farmland data view, and can simultaneously obtain the evidence chain generated by the proportional drift correction of this data.

[0109] Through the above processing, the system can generate self-calibration parameters for data sources, observation objects, and farmland spatiotemporal calibration units based on the deviation attribution results. It can then use these self-calibration parameters to perform traceable corrections on multi-source farmland raw data, extending data quality processing from anomaly identification to parameterized calibration and reliable view output. This improves the accuracy of farmland data correction, enhances the interpretability of calibration results, and improves the data reliability for subsequent agricultural decisions.

[0110] The following are system embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the system embodiments of this application, please refer to the method embodiments of this application.

[0111] Figure 2 This is a schematic diagram of the structure of the farmland multi-source data quality self-calibration system provided in the embodiments of this application. For example... Figure 2 As shown, the system includes: The acquisition module 201 is used to acquire multi-source farmland raw data of the target farmland and generate farmland spatiotemporal calibration units based on plot spatial information, agronomic management information and acquisition time information; Extraction module 202 is used to map multi-source farmland raw data to farmland spatiotemporal calibration units, extract quality features, and generate data source quality profiles; Module 203 is used to construct a farmland quality constraint map that characterizes the relationship between observation correlation, state evolution and quality transmission, based on farmland spatiotemporal calibration units, agronomic process constraints and data source quality profiles. The determination module 204 is used to determine the dynamic reliable anchor point based on the data source quality profile and farmland quality constraint map, and generate the corresponding self-calibration reference state; The identification module 205 is used to identify the type of data quality defect based on the cross-source consistency residuals between the observations from each data source and the self-calibration reference state, as well as the process constraint residuals between the observations from each data source and the agronomic process constraints. The generation module 206 is used to attribute the sources of deviations to the types of data quality defects based on the farmland quality constraint map and generate deviation attribution results. The correction module 207 is used to generate data source calibration parameters based on the deviation attribution results, correct the original multi-source farmland data using the data source calibration parameters, generate a reliable farmland data view, and update the data source quality profile, farmland quality constraint map and data source calibration parameters based on subsequent feedback.

[0112] In some embodiments, Figure 2 The acquisition module 201 receives multi-source farmland raw data carrying collection metadata according to the target farmland data access protocol, and parses the data source identifier, collection location, collection time and observation object from the collection metadata; it performs farmland object segmentation on the target farmland based on the plot spatial information to generate spatial calibration objects; it configures agronomic status identifiers for spatial calibration objects based on agronomic management information, and determines the corresponding time calibration window based on the collection time information; it binds the spatial calibration objects, agronomic status identifiers and time calibration windows to generate farmland spatiotemporal calibration units with data carrying relationships.

[0113] In some embodiments, Figure 2The extraction module 202 matches the observation values ​​of each data source to the corresponding farmland spatiotemporal calibration unit based on the collected metadata and observation objects carried by the original farmland data from multiple sources; it performs spatiotemporal benchmark unification and agronomic semantic alignment on the matched observation values ​​of each data source to generate standardized farmland observation data; based on the standardized farmland observation data, it extracts quality features that characterize the collection stability, observation consistency, spatiotemporal representativeness, and equipment operating status; based on the quality features, it determines the quality reliability and quality deviation characterization of each data source within the corresponding farmland spatiotemporal calibration unit, and generates a data source quality profile associated with the data source identifier, observation object, and farmland spatiotemporal calibration unit.

[0114] In some embodiments, Figure 2 The construction module 203 uses farmland spatiotemporal calibration units as map indexes to generate map nodes that characterize farmland status, data source observations, and data quality status. It converts agronomic process constraints into state evolution rules that connect different map nodes and determines the constraint relationships of farmland status changes with management and environmental processes based on the state evolution rules. It configures quality attributes for map nodes corresponding to each data source observation based on the data source quality profile and establishes observational correlations between data source observations and farmland status. Based on observational correlations and constraint relationships, it determines quality transmission parameters and generates farmland quality constraint maps to support the selection of reliable anchor points, residual analysis, and bias attribution.

[0115] In some embodiments, Figure 2 The determination module 204 determines the quality reliability of each data source within the corresponding farmland spatiotemporal calibration unit based on the data source quality profile; calculates the anchor point fit of each data source observation relative to the farmland state evolution relationship and observation correlation relationship according to the farmland quality constraint map; determines the data source observations that meet the anchor point selection conditions as dynamic reliable anchor points, and fuses the dynamic reliable anchor points according to the anchor point fit; when there are insufficient dynamic reliable anchor points, performs state inference based on the state evolution relationship in the farmland quality constraint map to generate the self-calibration reference state of the corresponding farmland spatiotemporal calibration unit.

[0116] In some embodiments, Figure 2 The determination module 204 determines the observation nodes corresponding to the observation values ​​of each data source and the associated farmland state nodes in the farmland quality constraint map; based on the state evolution relationship of the farmland state nodes, it calculates the evolution consistency between the observation values ​​of each data source and the current farmland state change trend; based on the observation correlation relationship of the observation nodes, it calculates the observation consistency between the observation values ​​of each data source and the associated observation values ​​within the same farmland spatiotemporal calibration unit; and based on the evolution consistency, observation consistency and the quality reliability of the corresponding data source, it generates the anchor point fit of the observation values ​​of each data source.

[0117] In some embodiments, Figure 2The identification module 205 compares the observations from each data source with the self-calibration reference state of the corresponding farmland spatiotemporal calibration unit to generate cross-source consistency residuals. Based on agronomic process constraints, it performs constraint verification on the state changes of the observations from each data source within the corresponding time calibration window to generate process constraint residuals. It integrates the cross-source consistency residuals, process constraint residuals, and data source quality profiles to generate a quality anomaly discrimination characterization. The quality anomaly discrimination characterization is input into the quality defect identification model to determine the data quality defect types associated with the data source, the observed object, and the farmland spatiotemporal calibration unit.

[0118] In some embodiments, Figure 2 The generation module 206 takes the abnormal observation node as the starting point for backtracking, searches for paths along the observation associations, state evolutions and quality transmission relationships in the farmland quality constraint map, and generates candidate deviation source paths; determines the corresponding source contribution based on the residual interpretation ability and quality transmission parameters of each candidate deviation source path; determines the target deviation source path from the candidate deviation source paths based on the source contribution, and generates deviation attribution results containing deviation source nodes, attribution paths and contribution characterization.

[0119] In some embodiments, Figure 2 The correction module 207 determines the deviation action mode and deviation impact range corresponding to the target data source based on the deviation attribution results; based on the deviation action mode, deviation impact range and data quality defect type, it generates data source calibration parameters associated with the data source, observation object and farmland spatiotemporal calibration unit; it performs observation value correction and confidence reassessment on the multi-source farmland raw data using the data source calibration parameters to generate calibrated farmland data; and it associates the calibrated farmland data with the corresponding raw observation values, data source calibration parameters, deviation attribution results and confidence characterization to generate a reliable farmland data view.

[0120] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0121] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various system embodiments described above.

[0122] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.

[0123] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0124] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

[0125] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0126] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which may be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0127] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for self-calibrating the quality of multi-source farmland data, characterized in that, include: Acquire multi-source raw farmland data of the target farmland, and generate farmland spatiotemporal calibration units based on plot spatial information, agronomic management information, and collection time information; The original farmland data from multiple sources is mapped to the farmland spatiotemporal calibration unit, quality features are extracted, and a data source quality profile is generated. Based on the farmland spatiotemporal calibration unit, agronomic process constraints, and data source quality profile, a farmland quality constraint map characterizing the relationship between observation correlation, state evolution, and quality transmission is constructed. Based on the data source quality profile and the farmland quality constraint map, determine the dynamic reliable anchor point and generate the corresponding self-calibration reference state; Based on the cross-source consistency residuals between the observations from each data source and the self-calibration reference state, and the process constraint residuals between the observations from each data source and the agronomic process constraints, the types of data quality defects are identified. Based on the farmland quality constraint map, the sources of deviations for the data quality defect types are attributed, and deviation attribution results are generated. The data source calibration parameters are generated based on the deviation attribution results. The original multi-source farmland data is corrected using the data source calibration parameters to generate a reliable farmland data view. The data source quality profile, the farmland quality constraint map, and the data source calibration parameters are updated based on subsequent feedback.

2. The method according to claim 1, characterized in that, The process of acquiring multi-source raw farmland data of the target farmland and generating a farmland spatiotemporal calibration unit based on plot spatial information, agronomic management information, and collection time information includes: According to the data access protocol of the target farmland, the raw farmland data from multiple sources carrying the collected metadata is received, and the data source identifier, collection location, collection time and observation object are parsed from the collected metadata. Based on the spatial information of the land parcel, the target farmland is divided into farmland objects to generate spatial calibration objects; Based on the agronomic management information, configure the agronomic status identifier for the spatial calibration object, and determine the corresponding time calibration window based on the acquisition time information; The spatial calibration object, agronomic status identifier, and time calibration window are bound together to generate a farmland spatiotemporal calibration unit with data carrying relationship.

3. The method according to claim 1, characterized in that, The step of mapping the multi-source farmland raw data to the farmland spatiotemporal calibration unit, extracting quality features, and generating a data source quality profile includes: Based on the collected metadata and observation objects carried by the multi-source farmland raw data, the observation values ​​of each data source are matched to the corresponding farmland spatiotemporal calibration unit; The spatiotemporal benchmarks of the matched data sources are unified and the agronomic semantics are aligned to generate standardized farmland observation data. Based on the standardized farmland observation data, quality characteristics that characterize the collection stability, observation consistency, spatiotemporal representativeness, and equipment operating status are extracted. Based on the quality characteristics, the quality reliability and quality deviation characterization of each data source within the corresponding farmland spatiotemporal calibration unit are determined, and a data source quality profile associated with the data source identifier, the observed object, and the farmland spatiotemporal calibration unit is generated.

4. The method according to claim 1, characterized in that, The process of constructing a farmland quality constraint map characterizing observational correlations, state evolution, and quality transmission relationships based on the farmland spatiotemporal calibration unit, agronomic process constraints, and data source quality profiles includes: Using the farmland spatiotemporal calibration unit as a map index, map nodes representing farmland status, data source observations, and data quality status are generated; The agronomic process constraints are converted into state evolution rules connecting different graph nodes, and the constraint relationships of farmland state changes with management and environmental processes are determined according to the state evolution rules. Based on the data source quality profile, quality attributes are configured for the map nodes corresponding to each data source observation, and an observation association is established between data source observations and farmland status. Based on the observed correlations and the constraint relationships, quality transmission parameters are determined, and a farmland quality constraint map is generated to support the selection of credible anchor points, residual analysis, and bias attribution.

5. The method according to claim 1, characterized in that, The step of determining dynamic reliable anchor points based on the data source quality profile and the farmland quality constraint map, and generating corresponding self-calibration reference states, includes: Based on the data source quality profile, the quality reliability of each data source within the corresponding farmland spatiotemporal calibration unit is determined. Based on the farmland quality constraint map, calculate the anchor point fit of each data source observation relative to the farmland state evolution relationship and observation correlation relationship; Data source observations that meet the anchor point selection criteria are identified as dynamic trusted anchor points, and these dynamic trusted anchor points are fused based on the anchor point fit. When the dynamic reliable anchor points are insufficient, state inference is performed based on the state evolution relationship in the farmland quality constraint map to generate the self-calibration reference state of the corresponding farmland spatiotemporal calibration unit.

6. The method according to claim 5, characterized in that, The step of calculating the anchor point fit of each data source observation relative to the farmland state evolution relationship and observation correlation relationship based on the farmland quality constraint map includes: In the farmland quality constraint map, determine the observation nodes corresponding to the observation values ​​of each data source and the associated farmland status nodes; Based on the state evolution relationship of the farmland state nodes, the evolution consistency between the observation values ​​of each data source and the current farmland state change trend is calculated; Based on the observation correlation relationship connected by the observation nodes, the observation consistency between the observation values ​​of each data source and the associated observation values ​​within the same farmland spatiotemporal calibration unit is calculated. Based on the evolutionary consistency, the observation consistency, and the quality reliability of the corresponding data source, an anchor fit degree is generated for the observations of each data source.

7. The method according to claim 1, characterized in that, The method of identifying data quality defect types based on the cross-source consistency residuals between observations from each data source and the self-calibration reference state, and the process constraint residuals between observations from each data source and the agronomic process constraints, includes: The observations from each data source are compared with the self-calibration reference state of the corresponding farmland spatiotemporal calibration unit to generate cross-source consistency residuals. Based on the agronomic process constraints, the state changes of each data source observation value within the corresponding time calibration window are constrained and verified to generate process constraint residuals. By integrating the cross-source consistency residual, the process constraint residual, and the data source quality profile, a quality anomaly discrimination characterization is generated. The quality anomaly discrimination characterization is input into the quality defect identification model to determine the data quality defect types associated with the data source, the observed object, and the farmland spatiotemporal calibration unit.

8. The method according to claim 1, characterized in that, The process of attributing the sources of deviations to the data quality defect types based on the farmland quality constraint map and generating deviation attribution results includes: Based on the data quality defect type, determine the corresponding abnormal observation node and associated quality status node in the farmland quality constraint map; Using the abnormal observation node as the starting point for backtracking, a path search is performed along the observation correlation, state evolution, and quality transmission relationship in the farmland quality constraint map to generate candidate deviation source paths; The contribution of each source is determined based on the residual explanation capability and quality transmission parameters of each candidate bias source path; Based on the source contribution, the target deviation source path is determined from the candidate deviation source paths, and a deviation attribution result containing deviation source nodes, attribution paths, and contribution characterizations is generated.

9. The method according to claim 1, characterized in that, The process of generating data sources based on the deviation attribution results using calibration parameters, correcting the original multi-source farmland data using the calibration parameters, and generating a reliable farmland data view includes: Based on the deviation attribution results, determine the deviation action mode and deviation impact range corresponding to the target data source; Based on the deviation action mode, the deviation influence range, and the data quality defect type, data sources associated with the data source, the observed object, and the farmland spatiotemporal calibration unit are generated from calibration parameters. The data source is used to perform observation correction and confidence reassessment on the original multi-source farmland data using calibration parameters to generate calibrated farmland data. The calibrated farmland data is correlated with the corresponding original observations, data sources, calibration parameters, bias attribution results, and confidence characterization to generate a reliable farmland data view.

10. A self-calibration system for the quality of multi-source farmland data, characterized in that, include: The acquisition module is used to acquire multi-source farmland raw data of the target farmland and generate farmland spatiotemporal calibration units based on plot spatial information, agronomic management information and collection time information; The extraction module is used to map the multi-source farmland raw data to the farmland spatiotemporal calibration unit, extract quality features, and generate a data source quality profile. The construction module is used to construct a farmland quality constraint map that characterizes the relationship between observation correlation, state evolution and quality transmission, based on the farmland spatiotemporal calibration unit, agronomic process constraints and the data source quality profile; The determination module is used to determine dynamic reliable anchor points based on the data source quality profile and the farmland quality constraint map, and generate corresponding self-calibration reference states; The identification module is used to identify the type of data quality defect based on the cross-source consistency residuals between the observations from each data source and the self-calibration reference state, and the process constraint residuals between the observations from each data source and the agronomic process constraints. The generation module is used to attribute the sources of deviations to the data quality defect types based on the farmland quality constraint map and generate deviation attribution results. The correction module is used to generate data source calibration parameters based on the deviation attribution results, correct the original multi-source farmland data using the data source calibration parameters, generate a reliable farmland data view, and update the data source quality profile, the farmland quality constraint map, and the data source calibration parameters based on subsequent feedback.

Images