Agricultural information real-time analysis method and system based on multi-source sensor data

Abstract

The application relates to the technical field of agricultural information processing, and discloses a real-time analysis method and system for agricultural information based on multi-source sensor data.The method comprises the following steps: collecting numerical data and visual data of a target farmland in real time through multi-source sensors; combining farmland agronomic conditions, analyzing numerical semantic events, identifying phenotypic semantic events, and merging the numerical semantic events and the phenotypic semantic events into multi-source semantic events; then analyzing the time sequence and position relationship of the events, dividing time sequence leading events and subsequent events to construct a cause-and-effect event pair, connecting the cause-and-effect event pair to form a cause-and-effect event chain, fusing the semantic event types of each node and the conduction process to generate farmland context features; finally, tracing back to the root semantic events and inducing factors, extracting agronomic attribution information along the conduction path of the cause-and-effect event chain, and comprehensively generating an agricultural information analysis report.The application aims to solve the problem that the prior art cannot clarify the cause-and-effect correlation and conduction logic of farmland abnormal events.

Description

Technical Field

[0001] This invention relates to the field of agricultural information processing technology, and in particular to a method and system for real-time analysis of agricultural information based on multi-source sensor data. Background Technology

[0002] With the rapid development of smart agriculture technology, farmland information collection and analysis technology based on multi-source sensors has been widely applied in agricultural production management. Current mainstream agricultural information processing technologies typically employ distributed sensor networks to collect environmental, soil, and crop growth data from farmland. After preliminary noise reduction and normalization processing of the raw data via edge computing nodes, the data is uploaded to a cloud server for centralized storage and statistical analysis, ultimately generating monitoring reports containing real-time values ​​of various parameters, historical change curves, and threshold alarm information. This type of technology enables automated collection and visualization of basic farmland production data, replacing the traditional manual field inspection mode and improving the efficiency of agricultural production management to a certain extent. It also provides data references for basic agricultural operations such as daily irrigation and fertilization.

[0003] However, in practical applications, existing analytical systems rely solely on threshold comparisons of single parameters or simple statistical correlation analysis to process multi-source data. They fail to combine the temporal and spatial relationships of agricultural events with the inherent agronomic transmission mechanisms of crop growth, thus failing to construct the causal transmission logic between agricultural events and making it difficult to form a systematic and logical understanding of the farmland production process. Summary of the Invention

[0004] This invention provides a method and system for real-time analysis of agricultural information based on multi-source sensor data. Its main purpose is to solve the problem that existing technologies cannot clarify the causal relationships and transmission logic of abnormal events in farmland.

[0005] To achieve the above objectives, the present invention provides a real-time agricultural information analysis method based on multi-source sensor data, comprising: S1: Real-time acquisition of environmental data, soil data, and crop image data of the target farmland through multi-source sensors, and encapsulation of the environmental data and soil data into numerical data, and crop image data into visual data; S2: Based on the agronomic conditions of the target farmland, analyze the numerical semantic events in the numerical data, identify the phenotypic semantic events in the visual data, and merge the numerical semantic events and the phenotypic semantic events into multi-source semantic events; S3: Analyze the temporal and positional relationships of the multi-source semantic events, divide the multi-source semantic events at the same position into temporal leading events and temporal following events, and construct causal event pairs of the multi-source semantic events based on the temporal leading events and the temporal following events; S4: Connect the causal event pairs with shared event nodes to form a causal event chain of the target farmland, and fuse the semantic event types and transmission processes of each node in the causal event chain to generate the contextual features of the target farmland; S5: Based on the transmission path of each semantic event type in the context feature, trace the node dependency relationship of the causal event chain, locate the root semantic event of the current context feature according to the node dependency relationship, and analyze the inducing factors of the context feature. S6: Based on the root semantic event and the triggering factor, extract the agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain, and generate an agricultural information analysis report of the target farmland by combining the contextual features and the agronomic attribution information.

[0006] Preferably, the step of analyzing numerical semantic events in the numerical data and identifying phenotypic semantic events in the visual data based on the agronomic conditions of the target farmland includes: By combining the agronomic conditions of the target farmland with the growth parameters of the crops in the target farmland, the numerical data is verified, and outlier data points are screened out from the numerical data. Based on the abnormal data points and the state of the target farmland, the numerical semantic events of the target farmland are obtained; Key phenotypic features are extracted from the visual data, compared with the normal growth state of crops in the target farmland, and feature variation points of the visual data are obtained. These feature variation points are then summarized as phenotypic semantic events.

[0007] Preferably, the step of analyzing the temporal and positional relationships of the multi-source semantic events, and dividing the multi-source semantic events at the same position into temporal leading events and temporal successor events, includes: Extract the location coordinates of the occurrence of the multi-source semantic events, and aggregate the multi-source semantic events with the same location coordinates into a spatially consistent event set. Sort the semantic events in the spatially consistent event set according to their occurrence time to obtain a semantic event sequence. The semantic topic dependencies of adjacent events in the semantic event sequence are evaluated to obtain the dependency evaluation results of the semantic event sequence. Based on the dependency evaluation results, the events in the semantic event sequence are divided to obtain the temporal leading events and temporal successor events.

[0008] Preferably, the step of constructing the causal event pair of the multi-source semantic event based on the temporal preceding event and the temporal succeeding event includes: Based on the agronomic mechanism of the target farmland, the transmission medium type of the temporal leading event and the temporal successor event in the same spatial location is verified to obtain the candidate causal events of the multi-source semantic event; By comparing the start time of the leading event with the start time of the subsequent events in the candidate causal events, and analyzing the time difference between the two and the rationality of the physical transmission rate of the transmission medium based on the comparison results, the temporally reasonable event pairs of the multi-source semantic events are obtained. The continuity of action and the directionality of response between the physical field influence boundary of the preceding event and the spatial region boundary gradient direction of the subsequent event in the temporally reasonable event pair are evaluated to obtain the causal event pair of the multi-source semantic event.

[0009] Preferably, the chaining of the causal event pairs with shared event nodes to form the causal event chain of the target farmland includes: The event nodes are sorted according to the chronological order of the events in the causal event pair to obtain the time node sequence of the causal event pair. Event nodes with empty inbound edge lists in the time node sequence are extracted as root candidate nodes. Starting from each of the root candidate nodes, trace subsequent events along the outgoing edge direction to obtain an initial causal chain segment. Locate the successor node with multiple incoming edge branches in the initial causal chain segment to obtain the branch convergence node of the initial causal chain segment. The agronomic contribution of each predecessor branch corresponding to the branch convergence node is compared to obtain the dominant branch and non-dominant branch corresponding to the branch convergence node. The non-dominant branch is merged into the link where the dominant branch is located to obtain a branchless merged causal chain segment. By connecting the spatiotemporal continuity between the end events and the beginning events of adjacent segments in the branchless merged causal chain segment, the causal event chain of the target farmland is obtained.

[0010] Preferably, the step of fusing the semantic event types and their transmission processes of each node in the causal event chain to generate the contextual features of the target farmland includes: Extract the propagation characteristics of the semantic event types of each node in the causal event chain to obtain the propagation characteristics of the semantic event types; By filtering the spatial density of the transmission process in the causal event chain, the key transmission path of the transmission process is obtained; By integrating the key transmission paths and propagation characteristics of each node in the causal event chain, the contextual features of the target farmland are generated.

[0011] Preferably, tracing the node dependencies of the causal event chain based on the propagation path of each semantic event type in the context features includes: Traverse the transmission paths in the context features in reverse order, and anchor the starting node and key interruption point of the transmission path; Based on the starting node, the continuity of semantic event types in the transmission path is analyzed one by one to obtain the dependency relationship between nodes in the transmission path and generate the node dependency relationship of the causal event chain.

[0012] Preferably, the step of locating the root semantic event of the current context feature based on the node dependency relationship and analyzing the triggering factors of the context feature includes: Analyze the dependency depth of each semantic event node in the node dependency relationship, and sort the semantic event nodes according to the dependency depth to obtain the event node sequence; The agronomic dominance of node event types in the event node sequence is analyzed, and combined with the spatial distribution aggregation degree assessment of the target farmland, the node with the largest dependency depth and the highest matching degree between the node event type and the agronomic conditions of the target farmland is identified as the root semantic event. Extract the region where the feature parameters of the root semantic event exceed the target crop growth standard parameters, and use the features of the region where the difference exceeds the target crop growth standard parameters as the inducing factors of the context features.

[0013] Preferably, the step of extracting agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain based on the root semantic event and the triggering factor includes: Locate the starting node of the root semantic event in the causal event chain, and determine the screening framework for agronomic attribution in the target farmland based on the semantic event type of the inducing factor and the agronomic conditions of the target farmland. Starting from the starting node, traverse along the propagation direction of the causal event chain to each direct successor node, and verify the spatiotemporal continuity between the direct successor node and the upstream node; Match the semantic event type with the filtering framework, and combine the agronomic mechanism of the target farmland to filter relevant influencing factors; The filtering results of all nodes in the combined sequence form the agronomic attribution information of the root semantic event.

[0014] To address the aforementioned problems, this invention also provides a real-time agricultural information analysis system based on multi-source sensor data, comprising: Data acquisition module: Collects environmental data, soil data and crop image data of the target farmland in real time through multi-source sensors, and encapsulates the environmental data and soil data into numerical data, and treats the crop image data as visual data; Semantic parsing module: Based on the agronomic conditions of the target farmland, it parses the numerical semantic events in the numerical data, identifies the phenotypic semantic events in the visual data, and merges the numerical semantic events and the phenotypic semantic events into multi-source semantic events; Event parsing module: Analyzes the temporal and positional relationships of the multi-source semantic events, divides the multi-source semantic events at the same position into temporal leading events and temporal following events, and constructs causal event pairs of the multi-source semantic events based on the temporal leading events and the temporal following events; Event fusion module: Connects the causal event pairs with shared event nodes to form a causal event chain of the target farmland, fuses the semantic event types and their transmission processes of each node in the causal event chain, and generates the contextual features of the target farmland; Cause tracing module: Based on the transmission path of each semantic event type in the context feature, trace the node dependency relationship of the causal event chain, locate the root semantic event of the current context feature according to the node dependency relationship, and analyze the triggering factors of the context feature; Report generation module: Based on the root semantic event and the triggering factor, extract the agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain, and generate an agricultural information analysis report of the target farmland by integrating the contextual features and the agronomic attribution information.

[0015] Beneficial effects This solution analyzes unified multi-source semantic events by combining farmland agronomic conditions, and then constructs causal event pairs by analyzing the temporal and positional relationships of these events, linking them into a complete causal chain. This transforms scattered heterogeneous data into a logically related sequence of agricultural events, enabling spatiotemporal constrained causal modeling of complex farmland production processes. It solves the problems of semantic fragmentation of multi-source data, high false alarm rates of abnormal events due to relying solely on statistical correlation analysis, and inability to accurately identify the true causes behind events in existing technologies. It has the advantages of more closely reflecting the actual production patterns of farmland and significantly improving the accuracy of abnormal early warnings. It can help farmers accurately locate the root cause of problems and avoid blindly taking agricultural measures. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a real-time agricultural information analysis method based on multi-source sensor data, provided in an embodiment of the present invention. Figure 2 This is a functional block diagram of an agricultural information real-time analysis system based on multi-source sensor data provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] This application provides a method and system for real-time analysis of agricultural information based on multi-source sensor data. The executing entity of the method and system includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method and system for real-time analysis of agricultural information based on multi-source sensor data can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0019] Reference Figure 1 The diagram shown is a flowchart illustrating a real-time agricultural information analysis method based on multi-source sensor data according to an embodiment of the present invention. In this embodiment, the real-time agricultural information analysis method based on multi-source sensor data includes: S1: Real-time acquisition of environmental data, soil data, and crop image data of the target farmland through multi-source sensors, and encapsulation of the environmental data and soil data into numerical data, and crop image data into visual data; Specifically, a multi-source sensor is a collection of various sensor devices with different data acquisition capabilities deployed in different spatial locations in the target farmland. It is used to simultaneously acquire multi-dimensional environmental, soil, and crop growth status information of the farmland and is the core hardware carrier of the agricultural information perception layer.

[0020] The target farmland is a specific agricultural production area designated by the user that requires real-time monitoring and analysis of agricultural information. Agricultural management data such as its geographical boundaries, crop varieties, and planting patterns must be collected and used only after obtaining user authorization.

[0021] Environmental data are quantifiable physicochemical parameters in the atmospheric environment of the target farmland that affect the growth and development of crops. These mainly include air temperature, relative humidity, light intensity, wind speed, wind direction, rainfall, and atmospheric carbon dioxide concentration.

[0022] Soil data refers to the physicochemical properties of the topsoil in the target farmland, mainly including soil temperature, soil volumetric water content, soil pH, soil electrical conductivity, soil organic matter content, and soil nitrogen, phosphorus, and potassium available nutrient content.

[0023] Crop image data is visible light or near-infrared image data of crops in the target farmland acquired by visual sensors. It includes intuitive phenotypic information such as crop height, leaf area index, leaf color, disease and pest spots, and fruit morphology.

[0024] Numerical data is a structured digital dataset formed by encapsulating environmental and soil data after standardization of format, unification of units, and timestamp marking. It has the characteristics of being computable, storable, and transmissible.

[0025] Visual data consists of crop image data that retains the original pixel matrix format and spatial coordinate information without undergoing structured numerical transformation. It is specifically used for subsequent crop phenotypic feature extraction and visual semantic recognition.

[0026] Furthermore, firstly, through multi-source sensors, environmental data, soil data, and crop image data of the target farmland are acquired synchronously according to a preset acquisition frequency. Among them, environmental sensors are deployed at different heights above the farmland canopy to accurately capture changes in the atmospheric environment, soil sensors are buried at different depths in the crop root distribution layer to obtain layered soil information, and visual sensors are deployed between the farmland rows or on high-altitude platforms to cover the entire crop growth area.

[0027] After collection, the acquired environmental and soil data are standardized and packaged, and the data storage format and physical units of measurement are unified. Each data point is appended with a corresponding collection timestamp and precise spatial location coordinates to generate structured numerical data.

[0028] Meanwhile, the collected crop image data is not subjected to structured numerical transformation. Instead, it is simply appended with the collection timestamp and spatial location coordinates and directly used as visual data storage, providing two types of differentiated basic data for subsequent multi-source semantic event analysis.

[0029] For example, taking the process of multi-source semantic event parsing of a target farmland to be monitored after a certain planting entity in a certain region completes user authorization as an example, multi-source sensors are deployed in a grid of 10 meters × 10 meters in the designated target farmland. Each grid node is equipped with 1 environmental sensor, 1 soil sensor and 1 vision sensor. After all sensors complete time synchronization calibration, the collection frequency of environmental data and soil data is set to once every 15 minutes, and the collection frequency of crop image data is once every hour.

[0030] The system receives raw data transmitted from various sensors in real time, standardizes the format and units of environmental and soil parameters such as air temperature and soil moisture content, adds corresponding acquisition timestamps and grid node coordinates to each data point, and encapsulates it to generate numerical data; for crop visible light images acquired by visual sensors, only acquisition timestamps and grid node coordinates are added, and the original pixel matrix format is retained as visual data, thus completing the data acquisition and preprocessing operations of this step.

[0031] S2: Based on the agronomic conditions of the target farmland, analyze the numerical semantic events in the numerical data, identify the phenotypic semantic events in the visual data, and merge the numerical semantic events and the phenotypic semantic events into multi-source semantic events; In this embodiment: the step of analyzing numerical semantic events in the numerical data and identifying phenotypic semantic events in the visual data based on the agronomic conditions of the target farmland includes: By combining the agronomic conditions of the target farmland with the growth parameters of the crops in the target farmland, the numerical data is verified, and outlier data points are screened out from the numerical data. Based on the abnormal data points and the state of the target farmland, the numerical semantic events of the target farmland are obtained; Key phenotypic features are extracted from the visual data, compared with the normal growth state of crops in the target farmland, and feature variation points of the visual data are obtained. These feature variation points are then summarized as phenotypic semantic events.

[0032] Specifically, the agronomic conditions of the target farmland refer to the basic production factors and environmental constraints that the target farmland possesses in the agricultural production process, including the farmland's soil type, topographic slope, irrigation method, fertilization system, planting system, historical pest and disease occurrence patterns, and local climate zoning characteristics, etc., which are the core benchmarks for the semantic analysis of agricultural information.

[0033] Numerical semantic events refer to abnormal state events extracted from numerical data that have clear agronomic meanings. They serve as an intermediate carrier for converting abstract numerical values ​​into understandable agricultural production events.

[0034] Phenotypic semantic events refer to phenotypic feature events identified from visual data that reflect abnormal crop growth status. They serve as an intermediate carrier for converting visual image features into understandable agricultural production events.

[0035] Multi-source semantic events refer to a unified set of semantic events obtained by merging numerical semantic events and phenotypic semantic events at the same spatiotemporal location. This integrates the quantitative information of numerical data with the intuitive phenotypic information of visual data, achieving semantic fusion of multi-source heterogeneous data.

[0036] The growth parameters of crops in the target farmland refer to the standard growth indicators of specific crop varieties planted in the target farmland at different growth stages, including plant height range, leaf area index range, leaf chlorophyll content range, root depth range, nutrient requirement threshold, etc., which are provided by the agricultural knowledge base and dynamically updated according to the crop growth period.

[0037] In anomalous data point index-type data, data points that exceed the normal growth parameter range of the corresponding crop growth stage and do not conform to the natural fluctuation pattern under farmland agronomic conditions are the core basis for triggering numerical semantic events.

[0038] Key phenotypic features refer to the core image features in visual data that can directly reflect the growth and health status of crops, including leaf color features, leaf texture features, plant morphology features, lesion shape and size features, fruit development features, etc., which are extracted through computer vision algorithms.

[0039] The normal growth status of crops in the target farmland refers to the standard phenotypic state that a specific crop variety planted in the target farmland should exhibit at its current growth stage under suitable environmental and soil conditions.

[0040] Feature variation points refer to feature regions where there are significant differences between the key phenotypic features extracted from visual data and the standard features of normal crop growth. They are the core basis for triggering phenotypic semantic events.

[0041] Furthermore, the agronomic conditions of the target farmland are first matched with the growth parameters of the corresponding crops to construct a normal fluctuation range of numerical data for the current farmland. The collected numerical data are then compared and verified against this normal fluctuation range one by one. After eliminating invalid data caused by sensor failure, abnormal data points are selected. Combining the parameter type, numerical deviation degree, and real-time production status of the abnormal data points, they are converted into numerical semantic events with clear agronomic meaning.

[0042] Meanwhile, key phenotypic features of crops are extracted from visual data using computer vision feature extraction algorithms. The extracted key phenotypic features are then compared dimensionally with the standard features of the normal growth state of the crop at its current growth stage stored in the agricultural knowledge base. Significantly different feature variation points are identified, and these feature variation points are summarized as phenotypic semantic events according to their corresponding agronomic meanings.

[0043] Finally, numerical semantic events and phenotypic semantic events collected at the same time and in the same spatial location are associated and merged to form multi-source semantic events that integrate quantitative environmental soil information and intuitive crop phenotypic information, providing a unified semantic input for subsequent causal relationship analysis.

[0044] The computer vision feature extraction algorithms used in this solution are a collection of existing technologies widely used in the field of agricultural phenotyping. They integrate traditional manual feature extraction methods with deep learning feature extraction methods. The traditional methods include the HSV visual data color histogram algorithm for extracting leaf color features, the local binary pattern algorithm and gray-level co-occurrence matrix algorithm for extracting leaf texture features, and edge detection and morphological operation algorithms for extracting plant morphology and lesion contour features. The deep learning method uses a lightweight convolutional neural network model that has been pre-trained on a general image dataset and then fine-tuned on a publicly available agricultural crop phenotyping dataset. It can automatically extract discriminative low-level visual features and high-level semantic features from crop visible light or near-infrared images. It also has robust optimization for typical scenarios such as complex light changes in farmland, overlapping and occlusion of leaves, and background interference. All of the above algorithms are mature existing technologies well known to those skilled in the art. They can be flexibly selected and combined and directly deployed according to specific crop types and phenotyping recognition needs.

[0045] For example, consider the process of parsing multi-source semantic events of a target farmland to be monitored after a planting entity in a certain region has completed user authorization.

[0046] The system receives raw data transmitted from various sensors in real time, standardizes the format and units of environmental and soil parameters such as air temperature and soil moisture content, adds corresponding acquisition timestamps and grid node coordinates to each data point, and encapsulates it to generate numerical data; for crop visible light images acquired by visual sensors, only acquisition timestamps and grid node coordinates are added, and the original pixel matrix format is retained as visual data.

[0047] The system loads the agronomic conditions and crop growth parameters of the target farmland, constructs the normal fluctuation range of numerical data for the current growth stage, compares each numerical data point with this range, and filters out abnormal data points with low soil volumetric water content collected multiple times in a certain grid node. Combining the irrigation system and recent rainfall of the farmland, it converts these data points into a numerical semantic event of mild soil drought.

[0048] Meanwhile, the leaf color and texture features in the visual data corresponding to the grid node are extracted by computer vision algorithms and compared with the standard features of the normal growth state of the crop at the current growth stage. The characteristic variation points of yellowing at the leaf edge are identified and summarized as the phenotypic semantic event of yellowing at the leaf edge of the crop.

[0049] The system merges the soil mild drought numerical semantic events and crop leaf edge yellowing phenotypic semantic events at the same time and the same grid node to generate multi-source semantic events for that grid node.

[0050] S3: Analyze the temporal and positional relationships of the multi-source semantic events, divide the multi-source semantic events at the same position into temporal leading events and temporal following events, and construct causal event pairs of the multi-source semantic events based on the temporal leading events and the temporal following events; In this embodiment: the analysis of the temporal and positional relationships of the multi-source semantic events, dividing the multi-source semantic events at the same position into temporal leading events and temporal successor events, includes: Extract the location coordinates of the occurrence of the multi-source semantic events, and aggregate the multi-source semantic events with the same location coordinates into a spatially consistent event set. Sort the semantic events in the spatially consistent event set according to their occurrence time to obtain a semantic event sequence. The semantic topic dependencies of adjacent events in the semantic event sequence are evaluated to obtain the dependency evaluation results of the semantic event sequence. Based on the dependency evaluation results, the events in the semantic event sequence are divided to obtain the temporal leading events and temporal successor events.

[0051] Specifically, the location coordinates of a multi-source semantic event refer to the two-dimensional planar coordinate information of the sensor deployment node corresponding to the multi-source semantic event in the geographic space of the target farmland. The coordinates are synchronously acquired by the positioning module built into the sensor during data collection, accurately corresponding to the physical location of the preset grid node in the farmland, and used to uniquely identify the specific spatial area where the event occurred.

[0052] A spatially consistent event set refers to a set of events formed by aggregating all multi-source semantic events with identical location coordinates. All events in the set occur at the same physical grid node of the target farmland, eliminating interference from events in different spatial locations and ensuring spatial consistency in subsequent time series analysis.

[0053] A semantic event sequence is an ordered list of events obtained by arranging all multi-source semantic events within a spatially consistent event set in ascending order according to their respective collection timestamps. It fully presents the temporal sequence of agricultural-related events occurring at different points in time within the same spatial location. Semantic topic dependence refers to the degree of correlation between two adjacent multi-source semantic events in a semantic event sequence on an agronomic topic, reflecting whether the occurrence of the preceding event has a potential agronomic impact on the generation of the following event.

[0054] Dependency assessment results refer to the set of results obtained after analyzing the semantic topic dependency of each pair of adjacent events in a semantic event sequence. It includes the degree of dependency of each pair of adjacent events and the corresponding unique event identifier, and is used to determine the possibility of temporal causal relationship between events.

[0055] A temporal leading event refers to a multi-source semantic event that occurs earlier in a semantic event sequence and has a significant semantic topic dependence on subsequent events; it is the causal event in a potential causal relationship.

[0056] A temporal successor event refers to a multi-source semantic event in a semantic event sequence that occurs later than the temporal leader event and has a significant semantic topic dependency on the leader event. It is the result event in a potential causal relationship.

[0057] Furthermore, the location coordinate parameters of each multi-source semantic event are first extracted in batches from the generated multi-source semantic event database. All multi-source semantic events are grouped and aggregated using the location coordinates as unique identifiers. Events with perfectly matching coordinates are grouped into the same group, forming multiple independent spatially consistent event sets.

[0058] For each spatially consistent event set, the collection timestamp parameter of each event within the set is extracted. The events are then fully sorted according to the ascending order of timestamps to generate a semantic event sequence corresponding to the spatial location. Subsequently, semantic topic matching analysis is performed on each pair of adjacent events in the semantic event sequence to obtain the semantic topic dependency of each pair of adjacent events, and the results are summarized to form a complete dependency evaluation result.

[0059] Finally, the dependency assessment results are compared with the preset dependency threshold. Adjacent event pairs with a dependency level higher than the threshold are divided, with the earlier event being marked as the temporal leader event and the later event being marked as the temporal successor event, thus completing the temporal division operation of multi-source semantic events at the same location.

[0060] Among them, the preset dependency threshold is essentially a quantitative boundary value used to determine whether there is a potential agronomic causal relationship between adjacent multi-source semantic events. It is not a fixed artificial preset constant, but is dynamically obtained in the following way: First, based on the historical event association data of the target farmland with user authorization for the past complete crop growth cycle, the semantic topic dependency degree distribution of various verified causal event pairs under the same crop variety, the same growth stage, and the same spatial grid location is statistically analyzed, and the lower limit of the high confidence interval is taken as the initial benchmark threshold.

[0061] Subsequently, based on the basic agronomic conditions of the target farmland, such as soil type, irrigation system, and fertilization level, as well as environmental parameters such as real-time monitored air temperature, soil moisture, and light intensity, agronomic correction coefficients are introduced to adjust the initial threshold by weighting. Then, the threshold parameters are continuously iteratively optimized based on the real causal events verified by subsequent agricultural operations.

[0062] For target farmland that is being monitored for the first time and has no local historical data, the general crop event association benchmark threshold with the same climate zone and planting pattern is first adopted, and rapid calibration is completed by collecting local data within the first growth cycle, so as to obtain the dependency threshold that is adapted to the actual production status of the current farmland.

[0063] For example, taking the target farmland designated for monitoring by a certain planting entity in a certain region after completing user authorization as an example, firstly, the location coordinates of all generated multi-source semantic events are extracted, and the soil mild drought numerical semantic events, crop leaf edge yellowing phenotypic semantic events, and crop leaf wilting phenotypic semantic events collected later at the same grid node are aggregated into a spatially consistent event set for that grid node.

[0064] Extract the collection timestamps of three events within the spatially consistent event set, and arrange them in chronological order to obtain the semantic event sequence of the node. The sequence order is as follows: mild soil drought event, yellowing of crop leaf edges event, and wilting of crop leaves event.

[0065] By loading an agricultural semantic topic association knowledge base, the semantic topic dependencies between mild soil drought events and crop leaf edge yellowing events, and between crop leaf edge yellowing events and crop leaf wilting events, are analyzed to obtain the degree of dependency between the two sets of events and form corresponding dependency assessment results.

[0066] Historical event data of the crop jointing stage in the past three growth cycles of the grid node were retrieved, and the average dependence of soil drought and leaf yellowing event pairs and leaf yellowing and leaf wilting event pairs were statistically obtained. The data were then corrected based on the current water sensitivity of the crop jointing stage, and the dependence thresholds of the two event pairs were obtained.

[0067] The obtained dependence levels were compared with the corresponding benchmark values, confirming that the dependence levels of both sets of events were higher than the benchmark values. Based on this, the semantic event sequence was divided, with the mild soil drought event marked as the temporal leading event of the crop leaf edge yellowing event, and the crop leaf edge yellowing event marked as the temporal successor event of the mild soil drought event. At the same time, the crop leaf edge yellowing event was marked as the temporal leading event of the crop leaf wilting event, and the crop leaf wilting event was marked as the temporal successor event of the crop leaf edge yellowing event.

[0068] In this embodiment: the step of constructing the causal event pair of the multi-source semantic event based on the temporal leader event and the temporal successor event includes: Based on the agronomic mechanism of the target farmland, the transmission medium type of the temporal leading event and the temporal successor event in the same spatial location is verified to obtain the candidate causal events of the multi-source semantic event; By comparing the start time of the leading event with the start time of the subsequent events in the candidate causal events, and analyzing the time difference between the two and the rationality of the physical transmission rate of the transmission medium based on the comparison results, the temporally reasonable event pairs of the multi-source semantic events are obtained. The continuity of action and the directionality of response between the physical field influence boundary of the preceding event and the spatial region boundary gradient direction of the subsequent event in the temporally reasonable event pair are evaluated to obtain the causal event pair of the multi-source semantic event.

[0069] Specifically, the agronomic mechanism of the target farmland refers to the intrinsic interaction patterns among various physiological and ecological events during the growth process of the corresponding crop varieties in the target farmland, which are obtained with user authorization. This includes the influence mechanisms of environmental factors such as water, nutrients, and temperature on crop growth, as well as the signal transduction patterns between different organs of the crop.

[0070] The type of transmission medium refers to the physical carrier through which causal effects between agricultural events are transmitted, including soil, air, water, crop roots, and crop vascular bundles.

[0071] Candidate causal events refer to a combination of a time-series leader event and a time-series successor event that, after verification by the type of transmission medium, has a potential causal relationship consistent with agronomic mechanisms.

[0072] The start time of a leading event refers to the time point at which the multi-source semantic event corresponding to the temporal leading event is first detected.

[0073] The start time of a subsequent event refers to the time point at which the multi-source semantic event corresponding to the sequential successor event is first detected.

[0074] The time difference is the absolute value of the time interval between the start time of the subsequent event and the start time of the preceding event, which is obtained by subtracting the two start times.

[0075] The physical transport rate of a conductive medium refers to the speed at which agricultural influencing factors propagate in the corresponding conductive medium, such as the infiltration rate of water in the soil and the transport rate of nutrients in the vascular bundles of crops.

[0076] A temporally reasonable event pair refers to a candidate causal event pair whose causal transmission time conforms to physical laws after verification of the reasonableness of the time difference and the physical transmission rate of the transmission medium.

[0077] The physical field influence boundary of a leading event refers to the maximum spatial range that the physical influence generated by the leading event can cover, as well as the distribution of the influence intensity within that range, which is obtained through analysis based on event type and intensity.

[0078] The spatial boundary of a subsequent event refers to the spatial range in which the subsequent event actually occurs and has an impact.

[0079] The boundary gradient direction refers to the direction in which the influence intensity of the physical field of the leading event decays from the central region to the edge region, and is obtained by gradient analysis of the physical field intensity distribution.

[0080] Continuity of action refers to whether the physical field influence of an event exhibits a continuous change in intensity within the spatial region of subsequent events. This is assessed by comparing the intensity values ​​of the physical field at different locations within that region.

[0081] Response directionality refers to whether the location and direction of subsequent events are consistent with the gradient direction of the physical field of the preceding event. It is evaluated by comparing the angle between the boundary of the spatial region of the subsequent event and the gradient direction of the physical field.

[0082] A causal event pair refers to a combination of a leading event and a following event that simultaneously satisfy the requirements of continuity of action and directionality of response and have a clear spatiotemporal causal relationship.

[0083] Furthermore, we first retrieve the agronomic mechanism knowledge corresponding to the target farmland authorized by the user from the agricultural knowledge base to clarify the possible transmission medium types between different types of agricultural events. Then, we extract the event type and spatial location information of all time-series leading events and time-series following events, and verify one by one whether there is a transmission medium that conforms to the agronomic mechanism in the same spatial location for each pair of time-series leading events and time-series following events. If there is, we mark the event pair as a candidate causal event pair. Finally, we summarize all the event pairs that meet the conditions to form a set of candidate causal event pairs.

[0084] First, extract the start times of the preceding and subsequent events in each candidate causal event pair, and analyze the time difference between the two times. Then, retrieve the physical transmission rate of the corresponding transmission medium from the agronomic physical parameter database. Combine the influence intensity and transmission distance of the preceding event to analyze the theoretical time range required for the causal transmission process, and determine whether the actual time difference falls within the theoretical time range. If it falls within the range, mark the candidate causal event pair as a temporally reasonable event pair. Finally, summarize all event pairs that meet the conditions to form a set of temporally reasonable event pairs.

[0085] First, for each temporally appropriate event pair, based on the type, intensity, and occurrence time of the preceding event, the corresponding physics model is invoked to analyze and obtain the physics field influence boundary and boundary gradient direction of the preceding event. Then, the spatial region boundary of the subsequent event is extracted, and the intensity change of the physics field of the preceding event in the spatial region of the subsequent event is analyzed to determine whether the spatial region boundary of the subsequent event is consistent with the gradient direction of the physics field of the preceding event. If both the continuity of action and the directionality of response are satisfied, the temporally appropriate event pair is marked as a causal event pair. Finally, all event pairs that meet the conditions are summarized to form the final multi-source semantic event causal event pair set.

[0086] Among them, the agronomic physical parameter database is a database that quantitatively stores the physical properties of agricultural transport media. It includes measurable parameters such as soil moisture infiltration rate and crop vascular bundle nutrient transport rate, providing benchmark data for verifying the spatiotemporal rationality of event transmission. This database is mainly obtained by analyzing collected environmental and soil dynamic data in combination with historical data of the target farmland.

[0087] For example: First, retrieve the agronomic mechanism knowledge corresponding to the target farmland, and verify that the transmission medium between soil moisture changes and crop leaf yellowing is the crop root system, which is consistent with the agronomic mechanism. Therefore, mark this event pair as a candidate causal event pair.

[0088] Next, the start time of the mild soil drought event and the start time of the yellowing of crop leaf edges were extracted to obtain the time difference between the two. The physical transmission rate of the crop root system conducting water stress signal was retrieved from the agronomic physical parameter database. Combined with the root depth and leaf position of the crop variety, the theoretical transmission time range was obtained. The actual time difference fell within this range. Therefore, the candidate causal event pair was marked as a temporally reasonable event pair.

[0089] Finally, based on the intensity and timing of the mild soil drought event, the physical field influence boundary of the event was analyzed to cover the entire grid node. The boundary gradient direction extended from the deep soil layer to the aboveground part of the crop. The spatial boundary of the yellowing event at the edge of the crop leaf was extracted as the crop leaf area within the grid node. The analysis showed that the intensity change of the soil moisture physical field in this area was continuous, and the location of the yellowing of the crop leaf was consistent with the direction of the physical field gradient. Therefore, this temporally reasonable event pair was marked as a causal event pair, completing the construction of this causal event pair.

[0090] S4: Connect the causal event pairs with shared event nodes to form a causal event chain of the target farmland, and fuse the semantic event types and transmission processes of each node in the causal event chain to generate the contextual features of the target farmland; In this embodiment: the chaining of causal event pairs with shared event nodes to form the causal event chain of the target farmland includes: The event nodes are sorted according to the chronological order of the events in the causal event pair to obtain the time node sequence of the causal event pair. Event nodes with empty inbound edge lists in the time node sequence are extracted as root candidate nodes. Starting from each of the root candidate nodes, trace subsequent events along the outgoing edge direction to obtain an initial causal chain segment. Locate the successor node with multiple incoming edge branches in the initial causal chain segment to obtain the branch convergence node of the initial causal chain segment. The agronomic contribution of each predecessor branch corresponding to the branch convergence node is compared to obtain the dominant branch and non-dominant branch corresponding to the branch convergence node. The non-dominant branch is merged into the link where the dominant branch is located to obtain a branchless merged causal chain segment. By connecting the spatiotemporal continuity between the end events and the beginning events of adjacent segments in the branchless merged causal chain segment, the causal event chain of the target farmland is obtained.

[0091] Specifically, the order in which events occur is determined by the order in which the timestamps attached to each event node are collected. The timestamps are marked in real time by the time synchronization module during the data acquisition from the multi-source sensors, accurate to the second.

[0092] An event node refers to each multi-source semantic event contained in a causal event pair. As the basic node unit in the causal relationship network, each node uniquely corresponds to an agricultural-related event that occurs at a specific spatiotemporal location.

[0093] The time node sequence is an ordered list formed by arranging all independent event nodes in ascending order according to their collection timestamps. It is generated by sorting the timestamps of all event nodes in the causal event set.

[0094] The inbound edge list is a set of all causal edges pointing to the current event node in a causal relationship network. Each causal edge corresponds to the connection relationship between the preceding event and the succeeding event in a causal event pair.

[0095] Root candidate nodes refer to event nodes whose incoming edge list is empty, that is, event nodes that have no other events as their leading causes. They are obtained by traversing the incoming edge list of each node in the time node sequence and filtering out nodes with 0 incoming edges.

[0096] Outgoing edge direction is the direction of the causal edge from the current event node to its subsequent event node in a causal relationship network, corresponding to the propagation direction from the preceding event to the subsequent event in a causal event pair.

[0097] The initial causal chain segment is a linear or branching link segment that starts from a single root candidate node and connects all reachable subsequent event nodes in sequence along the outgoing edge direction. It is generated by depth-first traversal of the causal relationship network.

[0098] An incoming edge branch consists of multiple different causal edges pointing to the same successor node, with each incoming edge corresponding to a different predecessor event node.

[0099] A branch convergence node is a successor event node with two or more incoming edge branches. It is obtained by traversing the incoming edge list of each node in the initial causal chain segment and filtering nodes with more than or equal to 2 incoming edges.

[0100] Agronomic contribution is a numerical indicator that quantifies the impact of each precursor branch on the occurrence of events at the branch convergence node. The indicator is obtained by comprehensive analysis of the agronomic mechanism, event intensity, and spatiotemporal matching degree of the target farmland.

[0101] The dominant branch is the predecessor branch with the highest agronomic contribution value, which is determined by comparing the agronomic contribution values ​​of all predecessor branches corresponding to the branch convergence node.

[0102] Non-dominant branches refer to all other predecessor branches whose agronomic contribution values ​​are lower than those of the dominant branch. They are determined by comparing the agronomic contribution values ​​of all predecessor branches corresponding to the branch convergence node.

[0103] A branchless merged causal chain segment is a branchless linear link segment formed by merging non-dominant branches into the dominant branch. It is generated by inserting all event nodes of the non-dominant branches into their corresponding positions in the dominant branch in chronological order.

[0104] Spatiotemporal continuity refers to the property that, in two adjacent unbranched causal chain segments, the time interval between the end time of the last event of the preceding segment and the start time of the first event of the following segment, as well as the spatial distance between the locations of the two events, conforms to the agronomic transmission law. The time interval parameter is represented by the timestamp of the event node.

[0105] Furthermore, firstly, all independent event nodes are extracted from all causal event pairs, and the collection timestamp associated with each event node is obtained. All event nodes are then fully sorted according to their collection timestamps in ascending order, generating a time node sequence covering all causal events. Subsequently, each event node in the time node sequence is traversed, and the incoming edge list of each node is checked one by one. The number of incoming edges for each node is counted, and event nodes with zero incoming edges are selected as root source candidate nodes. All selected root source candidate nodes are then aggregated to form the root source candidate node.

[0106] Each node in the root candidate nodes is selected as the starting node. Starting from the starting node, all reachable subsequent event nodes are visited sequentially along the outgoing edges in the causal relationship network using a depth-first traversal algorithm. All event nodes and causal edges passed during the traversal are recorded, and an initial causal chain segment corresponding to the starting node is generated.

[0107] During the process of traversing and generating all initial causal chain fragments, the length of the incoming edge list of each visited successor event node is checked synchronously. When the length of the incoming edge list of a successor node is detected to be greater than or equal to 2, the node is marked as a branch convergence node. All marked branch convergence nodes are summarized to form the branch convergence node set of the initial causal chain fragment.

[0108] For each branch convergence node, all predecessor branches corresponding to that node are extracted, and the agronomic contribution of each predecessor branch to the occurrence of the event at that branch convergence node is analyzed. Then, the agronomic contribution values ​​of all predecessor branches are sorted in descending order, and the predecessor branch with the largest value is selected as the dominant branch, while all other predecessor branches are designated as non-dominant branches.

[0109] Then, all event nodes contained in each non-dominant branch are inserted between the nodes at the corresponding time positions in the dominant branch according to their occurrence time order, completing the merging operation of non-dominant branches into the dominant branch, eliminating the branch structure in the initial causal chain segment, and generating a branchless merged causal chain segment.

[0110] All generated branchless merged causal chain fragments are sorted in ascending order according to the collection timestamp of their starting events, resulting in an ordered sequence of branchless merged causal chain fragments. Then, the spatiotemporal continuity of adjacent branchless merged causal chain fragments in the sequence is checked sequentially. This involves verifying whether the time interval between the end time of the last event of the preceding fragment and the start time of the starting event of the following fragment conforms to the agronomic transmission time range of the corresponding event type, and simultaneously verifying whether the spatial distance between the locations of the two events is within the influence range of the physical field of the corresponding event type.

[0111] For adjacent segments that meet the requirements of spatiotemporal continuity, the ending event of the preceding segment is connected to the starting event of the following segment through causal edges to complete the connection operation between segments. Finally, all the connected links are integrated into a complete causal event chain that covers all relevant causal events of the target farmland.

[0112] Among them, the Depth-First Search (DFS) algorithm's core principle is to start from a specified starting node and visit the nodes in the graph as deeply as possible along a continuous path until all nodes on the current path have been visited or there are no unvisited successor nodes. Then, it backtracks to the nearest node on the path with an unvisited branch and continues to explore other unvisited branch paths. The above process is repeated until all nodes in the graph connected to the starting node have been fully visited. This algorithm can be implemented through recursive calls or a stack data structure. It has the characteristics of low space complexity and natural path generation. It is suitable for path search, connectivity analysis, and subgraph extraction in graph structures. The above algorithms are all mature existing technologies well known to those skilled in the art. They can be flexibly selected and combined and directly deployed according to specific crop types and phenotypic recognition needs.

[0113] For example, taking the target farmland designated for monitoring by a certain planting entity in a certain region after completing user authorization as an example, four sets of causal event pairs have been generated through the preliminary steps, namely, mild soil drought event to crop leaf edge yellowing event, insufficient soil available nitrogen content event to crop leaf edge yellowing event, crop leaf edge yellowing event to crop leaf wilting event, and crop leaf wilting event to reduction in the number of effective ears per crop plant.

[0114] First, extract five independent event nodes from the above causal event pairs: mild soil drought, insufficient available nitrogen in soil, yellowing of crop leaf edges, wilting of crop leaves, and reduction in the number of effective ears per crop plant. Obtain the collection timestamp associated with each event node. Sort all event nodes in ascending order of collection timestamps to generate a time node sequence. The sequence order is as follows: mild soil drought, insufficient available nitrogen in soil, yellowing of crop leaf edges, wilting of crop leaves, and reduction in the number of effective ears per crop plant.

[0115] Traverse each event node in the time node sequence, check the incoming edge list of each node one by one, and count that the number of incoming edges for the soil mild drought event and the soil available nitrogen deficiency event are both 0. Mark these two nodes as root source candidate nodes and summarize them to form a root source candidate node set.

[0116] Secondly, the soil mild drought event node and the soil available nitrogen deficiency event node are selected sequentially from the root cause candidate node set as starting nodes. Starting from the two starting nodes respectively, all reachable subsequent event nodes are visited sequentially along the outgoing edges in the causal relationship network using a depth-first traversal algorithm. All event nodes and causal edges traversed during the traversal are recorded, generating two initial causal chain segments: The first initial causal chain segment is from a mild soil drought event to a yellowing of crop leaf margins event to a wilting of crop leaves event to a reduction in the number of effective ears per crop plant event; The second initial causal chain segment is from the event of insufficient available nitrogen in the soil to the event of yellowing of crop leaf edges to the event of wilting of crop leaves to the event of a reduction in the number of effective ears per crop plant.

[0117] During the process of traversing and generating two initial causal chain segments, the length of the incoming edge list of each visited successor event node is checked synchronously. When the length of the incoming edge list of the crop leaf edge yellowing event is detected to be 2, the node is marked as a branch convergence node.

[0118] Subsequently, for the crop leaf edge yellowing event at the branch convergence node, the precursor branches of the corresponding mild soil drought event and the soil available nitrogen deficiency event were extracted, and the agronomic contribution of the two precursor branches to the occurrence of crop leaf edge yellowing event was analyzed.

[0119] By combining the stress response data of the crop at the jointing stage in the historical production database of the target farmland authorized by the user, and the soil drought intensity and soil nitrogen deficiency intensity parameters obtained in this monitoring, the agronomic contribution of the precursor branch of the mild soil drought event and the agronomic contribution of the precursor branch of the insufficient soil available nitrogen content event were obtained.

[0120] The agronomic contribution values ​​of the two precursor branches are sorted in descending order. The precursor branch of the mild soil drought event is selected as the dominant branch, and the precursor branch of the insufficient soil available nitrogen content event is selected as the non-dominant branch. Then, the soil available nitrogen content insufficient event nodes contained in the non-dominant branch are inserted into the dominant branch between the mild soil drought event node and the crop leaf edge yellowing event node according to their occurrence time. This completes the merging operation of non-dominant branches into dominant branches, eliminates the branch structure in the initial causal chain fragment, and generates a branchless merged causal chain fragment.

[0121] Finally, the generated branchless merged causal chain fragments are arranged in ascending order according to the collection timestamp of their starting events, resulting in an ordered sequence of branchless merged causal chain fragments containing only one fragment. The spatiotemporal continuity of adjacent fragments in this sequence is checked, ultimately yielding the complete causal event chain corresponding to the target farmland. The order of events in the chain is as follows: mild soil drought, insufficient available nitrogen in the soil, yellowing of crop leaf edges, wilting of crop leaves, and reduction in the number of effective ears per crop plant.

[0122] In this embodiment: the step of fusing the semantic event types and their transmission processes of each node in the causal event chain to generate the contextual features of the target farmland includes: Extract the propagation characteristics of the semantic event types of each node in the causal event chain to obtain the propagation characteristics of the semantic event types; By filtering the spatial density of the transmission process in the causal event chain, the key transmission path of the transmission process is obtained; By integrating the key transmission paths and propagation characteristics of each node in the causal event chain, the contextual features of the target farmland are generated.

[0123] Specifically, a causal event chain is a complete linear link formed by connecting causal event pairs with shared event nodes, covering all relevant causal events of the target farmland. Each node uniquely corresponds to an agricultural-related event that occurs at a specific spatiotemporal location.

[0124] Semantic event types are classification identifiers for multi-source semantic events, divided into two categories: numerical semantic events and phenotypic semantic events. Numerical semantic events include anomalous state events parsed from numerical data, such as mild soil drought and insufficient available nitrogen content in the soil. Phenotypic semantic events include phenotypic anomalous events identified from visual data, such as yellowing of crop leaf edges, wilting of crop leaves, and reduction in the number of effective ears per crop plant.

[0125] Propagation characteristics are a set of spatiotemporal diffusion patterns of the impact of a semantic event type after it occurs in the target farmland. They include four core parameters: propagation rate, impact range attenuation coefficient, duration, and dependence on the transmission medium.

[0126] The transmission process is the complete process by which the influence of a preceding semantic event in a causal event chain is transmitted to subsequent semantic events through a specific transmission medium. Each transmission process corresponds to a causal edge in the causal event chain.

[0127] Spatial density is the frequency of the same conduction process occurring at different spatial locations within a unit area of ​​farmland. It is obtained by statistically analyzing the ratio of the number of occurrences of the corresponding conduction process in all grid nodes of the target farmland to the total area of ​​the target farmland. The basic data consists of spatial location information collected by multi-source sensors and spatial distribution data of multi-source semantic events.

[0128] The critical transmission path is an ordered sequence of transmission processes with the highest spatial density and the greatest weight in influencing the occurrence of subsequent events in the causal event chain. It is obtained by sorting the spatial density of all transmission processes in descending order and then filtering them.

[0129] Contextual features are a comprehensive set of features that integrate the propagation characteristics and key transmission paths of semantic event types at each node in the causal event chain. These features can fully reflect the current agricultural production status and event development trend of the target farmland. They include four core dimensions: event type, propagation pattern, scope of influence, and development trend. They are the core basis for subsequent causal tracing and analysis report generation.

[0130] Furthermore, the algorithm first traverses each event node in the causal event chain, extracts the semantic event type corresponding to each node, and retrieves the propagation rate, influence range attenuation coefficient, duration, and transmission medium dependence parameters corresponding to the semantic event type from the user-authorized target farmland agronomic mechanism knowledge base and historical production database. These parameters are then matched one by one with the occurrence time, spatial location, and event intensity of the event node to generate a set of propagation characteristics of each semantic event type under specific spatiotemporal conditions in the current target farmland.

[0131] Then, the transmission process corresponding to all causal edges in the causal event chain is extracted, the occurrence frequency of each transmission process in all grid nodes of the target farmland is counted, the occurrence frequency is divided by the total area of ​​the target farmland to obtain the spatial density of each transmission process, all transmission processes are sorted from high to low according to spatial density, the transmission processes with the highest spatial density are selected according to a preset proportion, and these transmission processes are connected in sequence according to their order in the causal event chain to form the key transmission path of the transmission process.

[0132] Finally, the propagation characteristics of each event node in the causal event chain are matched one-to-one with the propagation process passing through that node in the key transmission path. The propagation characteristics of each node are then systematically spliced ​​together according to the order of the key transmission path. At the same time, the spatial distribution information and temporal transmission information of the key transmission path are integrated to generate target farmland context features that include dimensions such as event type, propagation rate, scope of influence, and development trend.

[0133] Among them, the target farmland agronomic mechanism knowledge base is a database that stores the intrinsic physiological and ecological laws of crops in the target farmland. It covers the influence mechanism of environmental factors such as water, nutrients, and temperature on crop growth, as well as the signal transmission logic between crop organs. It provides users with agronomic benchmarks for analyzing multi-source semantic events. This knowledge base integrates historical production data and actual agronomic theories of the target farmland authorized by users, and stores them in a standardized growth response rule base according to agronomic conditions such as soil type and irrigation mode.

[0134] For example, taking a target farmland designated for monitoring by a certain planting entity in a certain region after user authorization as an example, the five nodes in the causal event chain are first traversed: mild soil drought event, insufficient available nitrogen content in soil event, yellowing of crop leaf edges event, wilting of crop leaves event, and reduction in the number of effective ears per crop plant event. The semantic event type corresponding to each node is extracted. From the agronomic mechanism knowledge base and historical production database of the target farmland authorized by the user, the propagation rate, attenuation coefficient of influence range, duration, and conduction medium parameters corresponding to the mild soil drought event, insufficient available nitrogen content in soil event, yellowing of crop leaf edges event, and wilting of crop leaves event are retrieved. The reduction in the number of effective ears per crop plant event has no subsequent propagation characteristics. These parameters are associated and matched with the occurrence time, spatial location, and event intensity of each event node to generate a set of propagation characteristics for each semantic event type.

[0135] Then, four transmission processes were extracted from the causal event chain: from mild soil drought to yellowing of crop leaf edges, from insufficient available nitrogen in the soil to yellowing of crop leaf edges, from yellowing of crop leaf edges to wilting of crop leaves, and from wilting of crop leaves to a reduction in the number of effective ears per crop plant. The frequency of occurrence of each transmission process in all grid nodes of the target farmland was counted, and the spatial density corresponding to each transmission process was analyzed. All transmission processes were sorted from high to low spatial density, and the transmission processes with the highest spatial density were selected and connected in sequence according to their order in the causal event chain to form the key transmission paths of the transmission process.

[0136] Finally, the propagation characteristics of each event node in the causal event chain are mapped one-to-one with the propagation process passing through that node in the key transmission path. The propagation characteristics of each node are then sequentially pieced together according to the order of the key transmission path. At the same time, the spatial distribution information and temporal transmission information of the key transmission path are integrated to generate target farmland context features that include dimensions such as event type, propagation pattern, scope of influence, and development trend. S5: Based on the transmission path of each semantic event type in the context feature, trace the node dependency relationship of the causal event chain, locate the root semantic event of the current context feature according to the node dependency relationship, and analyze the inducing factors of the context feature. In this embodiment: tracing the node dependencies of the causal event chain based on the propagation paths of each semantic event type in the context features includes: Traverse the transmission paths in the context features in reverse order, and anchor the starting node and key interruption point of the transmission path; Based on the starting node, the continuity of semantic event types in the transmission path is analyzed one by one to obtain the dependency relationship between nodes in the transmission path and generate the node dependency relationship of the causal event chain.

[0137] Specifically, the key interruption point is a semantic event node in the transmission path where the semantic event type changes abruptly or the transmission logic does not conform to the agronomic mechanism. It is obtained by comparing the semantic event types and transmission characteristics of adjacent nodes.

[0138] The continuity of semantic event types refers to the causal relationship between the semantic event types of adjacent semantic event nodes in the transmission path that conforms to the agronomic mechanism of the target farmland, and can form a complete event transmission logic.

[0139] The dependency relationship between nodes in the transmission path refers to the degree of causal dependence between adjacent semantic event nodes in the transmission path, which is obtained by analyzing the continuity of semantic event types and the transmission logic.

[0140] The node dependency relationship of a causal event chain refers to the set of dependency relationships between all semantic event nodes in the causal event chain, which is achieved by summarizing and integrating the node dependencies in all propagation paths.

[0141] Furthermore, firstly, all key transmission paths are extracted from the generated context features. Taking the terminal result node of each transmission path as the starting point of traversal, each semantic event node in the path is visited sequentially in the direction opposite to the normal transmission direction of causal events. During the traversal, the predecessor node identifier and successor node identifier of each node are recorded simultaneously. The semantic event type of each node is verified one by one to see if there is a logical relationship between the semantic event type of each node and the semantic event type of the adjacent nodes before and after it, which conforms to the agronomic mechanism of the target farmland.

[0142] When a semantic event node is detected to have no predecessor node, the node is marked as the starting node of the corresponding transmission path; when a pair of adjacent nodes is detected to have no semantic association that conforms to agronomic mechanisms, the second node in the pair of adjacent nodes is marked as the critical interruption point of the corresponding transmission path.

[0143] After completing the reverse traversal and node marking of all transmission paths, using all anchored starting nodes as the analysis benchmark, the entire transmission path corresponding to each starting node is traversed one by one. During the traversal, the semantic event types of adjacent nodes are compared node by node to verify whether the transmission of semantic event types from upstream nodes to downstream nodes meets the continuity requirement, to determine whether there is a direct causal dependency between adjacent nodes, and to quantify the degree of dependency between each pair of adjacent nodes based on the agronomic mechanism and event intensity parameters of the target farmland.

[0144] The dependency information of all adjacent node pairs in a single transmission path is summarized to form the node dependency relationship of that transmission path. Finally, the node dependency relationships of all transmission paths are integrated to generate a complete node dependency relationship covering the entire causal event chain.

[0145] For example, taking the target farmland designated for monitoring by a certain planting entity in a certain region after completing user authorization as an example, the transmission path in the generated context features is traversed in reverse. Taking the event of reduced effective ears per crop plant at the end node of the transmission path as the starting point of the traversal, the events of crop leaf wilting, crop leaf edge yellowing, insufficient available nitrogen content in soil, and mild soil drought are visited in sequence along the direction opposite to the transmission direction of causal events.

[0146] During the traversal, the predecessor node identifier and successor node identifier of each node are recorded. The agronomic logical association between the semantic event type of each node and the semantic event types of the adjacent nodes is verified one by one. When the mild soil drought event is detected to have no predecessor node, it is marked as the starting node of the transmission path. No semantic association interruption between adjacent nodes is detected, so no key interruption point is marked.

[0147] Based on the anchored starting node, mild soil drought event, the corresponding transmission paths are traversed one by one. The semantic event types of adjacent nodes are compared node by node to verify that the semantic transmission from mild soil drought event to insufficient soil available nitrogen content event, from insufficient soil available nitrogen content event to crop leaf edge yellowing event, from crop leaf edge yellowing event to crop leaf wilting event, and from crop leaf wilting event to reduced number of effective ears per crop plant event all conform to the agronomic mechanism of the target farmland and meet the continuity requirement of semantic event types. It is determined that there is a direct causal dependency relationship between adjacent nodes, and the node dependency relationship of the transmission path is summarized. The node dependency relationships of all transmission paths are integrated to generate a complete node dependency relationship covering the entire causal event chain.

[0148] In this embodiment: the step of locating the root semantic event of the current context feature based on the node dependency relationship and analyzing the inducing factors of the context feature includes: Analyze the dependency depth of each semantic event node in the node dependency relationship, and sort the semantic event nodes according to the dependency depth to obtain the event node sequence; The agronomic dominance of node event types in the event node sequence is analyzed, and combined with the spatial distribution aggregation degree assessment of the target farmland, the node with the largest dependency depth and the highest matching degree between the node event type and the agronomic conditions of the target farmland is identified as the root semantic event. Extract the region where the feature parameters of the root semantic event exceed the target crop growth standard parameters, and use the features of the region where the difference exceeds the target crop growth standard parameters as the inducing factors of the context features.

[0149] Specifically, a semantic event node is each multi-source semantic event contained in a causal event chain. As a basic node unit in the causal relationship network, each node uniquely corresponds to an agricultural-related event that occurs at a specific spatiotemporal location.

[0150] Dependency depth refers to the number of causal edges traversed from the starting node of the causal event chain to the current semantic event node. It is obtained by traversing the propagation path of the causal event chain in reverse and is used to quantify the hierarchical position of a node in the causal propagation process.

[0151] The event node sequence is an ordered list obtained by sorting all semantic event nodes according to their dependency depth.

[0152] Node event type is a classification identifier for multi-source semantic events, divided into two categories: numerical semantic events and phenotypic semantic events. Numerical semantic events include abnormal state events parsed from numerical data, such as mild soil drought and insufficient available nitrogen content in soil. Phenotypic semantic events include phenotypic abnormal events identified from visual data, such as yellowing of crop leaf edges and wilting of crop leaves.

[0153] Agronomic dominance is an indicator that quantifies the impact of a node event type on all subsequent events. It is derived from the agronomic mechanisms, event intensity, and transmission paths of the target farmland. The higher the value, the greater the impact of the event on subsequent events.

[0154] The spatial distribution aggregation degree of the target farmland is an indicator for evaluating the degree of concentration of the same type of semantic events in different spatial locations of the target farmland. It is obtained by statistically analyzing the occurrence frequency of this type of event at each grid node. The basic data is obtained by collecting spatial location information from multiple sources of sensors and spatial distribution data of multiple sources of semantic events.

[0155] Matching degree is the degree of fit between the node event type and the agronomic conditions of the target farmland. It is obtained by comparing the consistency between the influencing factors corresponding to the event type and the actual agronomic conditions of the target farmland. The higher the value, the more the event type matches the actual production environment of the target farmland.

[0156] The root semantic event is the most fundamental semantic event that leads to the generation of current contextual features. It is the node with the greatest dependency depth, the strongest agronomic dominance, and the highest degree of matching with the agronomic conditions of the target farmland in the causal event chain.

[0157] The feature parameters of a root semantic event are the quantifiable physicochemical parameters corresponding to the root semantic event. For example, the feature parameter of a mild soil drought event is soil volumetric water content, and the feature parameter of a soil available nitrogen deficiency event is soil available nitrogen content. These are raw numerical data collected by multi-source sensors.

[0158] The target farmland crop growth standard parameters are the standard growth indicators of specific crop varieties planted in the target farmland at different growth stages, including the suitable range of environmental parameters such as soil moisture, nutrients, and temperature.

[0159] The difference region is the numerical range or spatial region where the feature parameters of the root semantic event exceed the range of standard parameters for crop growth. It is obtained by comparing the actual feature parameters of the root semantic event with the standard parameters.

[0160] The triggering factors of contextual features are the specific features that lead to the occurrence of root semantic events.

[0161] Furthermore, we first traverse each semantic event node in the node dependency relationship, and then trace back all its predecessor nodes based on each node. We count the number of causal edges traversed from the starting node of the causal event chain to the current node, analyze the dependency depth of each semantic event node, and sort all semantic event nodes in descending order of dependency depth to generate an event node sequence containing all nodes.

[0162] Secondly, the node event type corresponding to the node is extracted. Combined with the agronomic mechanisms and event intensity data of the target farmland, the agronomic dominance of this node event type on all subsequent events is analyzed. Simultaneously, the frequency of this node event type across all grid nodes in the target farmland is statistically analyzed to obtain the spatial distribution aggregation degree of this event type. Then, the event type of each node is compared with the agronomic conditions of the target farmland to analyze their matching degree. Finally, based on the evaluation results of four dimensions—dependency depth, agronomic dominance, spatial distribution aggregation degree, and matching degree—the node with the largest dependency depth and the highest matching degree between its node event type and the agronomic conditions of the target farmland is identified as the root semantic event.

[0163] Finally, the actual feature parameters of the root semantic event are compared dimension-by-dimensionally with the corresponding growth standard parameters to identify the numerical ranges where the actual parameters exceed the standard parameter range and the corresponding spatial grid regions, forming difference regions. Finally, feature information such as the numerical values, spatial distribution range, and duration of feature parameters within these difference regions are extracted and used as inducing factors for the current contextual features.

[0164] For example, taking a target farmland in a certain region that a planting entity has authorized for monitoring in a specific area as an example, the process first iterates through the five semantic event nodes in the generated node dependency relationships: mild soil drought, insufficient available nitrogen content in soil, yellowing of crop leaf edges, wilting of crop leaves, and reduction in the number of effective ears per crop. Then, it traces back the predecessor node of each node, finding that the mild soil drought event has the highest dependency depth, while the reduction in the number of effective ears per crop has the lowest. Finally, all nodes are sorted according to their dependency depth from highest to lowest, resulting in the event node sequence: mild soil drought, insufficient available nitrogen content in soil, yellowing of crop leaf edges, wilting of crop leaves, and reduction in the number of effective ears per crop.

[0165] Then, each node in the event node sequence was traversed to extract the corresponding node event type. Combining this with the agronomic mechanisms and event intensity data of the target farmland, it was found that the mild soil drought event had the highest agronomic dominance, while the insufficient available nitrogen content event had a lower agronomic dominance. Simultaneously, the frequency of occurrence of each event type across all grid nodes in the target farmland was statistically analyzed, revealing a higher spatial distribution aggregation degree for the mild soil drought event and a lower spatial distribution aggregation degree for the insufficient available nitrogen content event. Comparing the event type of each node with the agronomic conditions of the target farmland, it was found that the mild soil drought event had the highest matching degree with the agronomic conditions of the target farmland, while the insufficient available nitrogen content event had a lower matching degree. After comprehensive evaluation, the mild soil drought event with the largest localization dependency depth and the highest matching degree was identified as the root semantic event.

[0166] Finally, the characteristic parameter of the root semantic event "slight soil drought"—soil volumetric water content—was extracted. The standard parameter range of soil volumetric water content for the target farmland's crops at the current growth stage was retrieved from an authorized agricultural knowledge base. The actual collected soil volumetric water content data was compared with the standard parameters to identify the numerical intervals where the actual soil volumetric water content was lower than the lower limit of the standard parameters and the corresponding spatial grid regions, forming difference regions. Feature information such as the numerical values, spatial distribution range, and duration of characteristic parameters within these difference regions was extracted, and this information was used as the inducing factors of the current contextual features.

[0167] S6: Based on the root semantic event and the triggering factor, extract the agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain, and generate an agricultural information analysis report of the target farmland by combining the contextual features and the agronomic attribution information.

[0168] In this embodiment: the step of extracting the agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain based on the root semantic event and the triggering factor includes: Locate the starting node of the root semantic event in the causal event chain, and determine the screening framework for agronomic attribution in the target farmland based on the semantic event type of the inducing factor and the agronomic conditions of the target farmland. Starting from the starting node, traverse along the propagation direction of the causal event chain to each direct successor node, and verify the spatiotemporal continuity between the direct successor node and the upstream node; Match the semantic event type with the filtering framework, and combine the agronomic mechanism of the target farmland to filter relevant influencing factors; The filtering results of all nodes in the combined sequence form the agronomic attribution information of the root semantic event.

[0169] Specifically, the agronomic attribution screening framework is a set of structured rules used to screen agronomic influencing factors related to root semantic events. It is constructed based on the semantic event type of the inducing factors and the agronomic conditions of the target farmland, and clarifies the dimensions of attribution information to be extracted under different semantic event types.

[0170] A direct successor node is a downstream node in a causal event chain that has a direct causal edge connection to the current node.

[0171] An upstream node is an upstream node in a causal event chain that has a direct causal edge connection with the current direct successor node, i.e., the preceding event node of the current direct successor node.

[0172] Spatiotemporal continuity is the property of whether the time interval and spatial distance between two adjacent event nodes conform to the laws of agronomic transmission.

[0173] Relevant influencing factors are environmental, soil, and crop physiological factors that are directly related to the occurrence and transmission of root semantic events in agronomical terms.

[0174] Agronomic attribution information is a structured collection of information that integrates all relevant influencing factors in the causal event chain, comprehensively reflecting the complete agronomic process of the root semantic event from its occurrence to its transmission to the final result.

[0175] Furthermore, firstly, within the constructed causal event chain, node retrieval is performed based on the unique identifier of the root semantic event to accurately locate the starting node corresponding to the root semantic event. The semantic event type corresponding to the inducing factor is obtained, and simultaneously, user-authorized agronomic condition data of the target farmland is retrieved. Combining the agronomic impact dimensions corresponding to the semantic event type of the inducing factor, as well as basic production factors of the target farmland such as soil type, irrigation method, and fertilization system, a structured agronomic attribution screening framework is constructed, including attribution dimensions, screening thresholds, and verification rules. This clarifies the scope and standards of various influencing factors to be extracted in subsequent steps.

[0176] Starting from the identified initial node, the process proceeds along the predefined event propagation direction in the causal event chain, sequentially visiting each direct successor node that has a direct causal connection to the initial node. For each visited direct successor node, the collection timestamp and location coordinates are extracted, along with the collection timestamp and location coordinates of its upstream node. The time interval and spatial distance between the two nodes are analyzed, and the results are compared with the agronomic propagation time range and physical field influence range of the corresponding semantic event type to verify whether the spatiotemporal continuity between them conforms to agronomic laws.

[0177] Extract the semantic event type corresponding to the currently accessed node, and match this semantic event type with the agronomic attribution screening framework constructed in the first step dimension by dimension to determine the categories of influencing factors that need to be screened under this node. Retrieve the agronomic mechanism data of the target farmland authorized by the user, and combine it with the causal transmission relationship between this node and upstream nodes. From the multi-source semantic event feature parameters corresponding to this node, screen out the relevant influencing factors that have a direct agronomical association with the occurrence and transmission of the root semantic event, and exclude irrelevant interference factors.

[0178] According to the order of nodes in the causal event chain, the relevant influencing factors of each node are collected in sequence after screening. The influencing factors of all nodes are combined in an orderly manner according to the logical order of event transmission to form a complete structured information set containing the initial characteristics of the root semantic event, the intermediate influencing factors of each transmission node, and the final result characteristics, that is, the agronomic attribution information of the root semantic event.

[0179] For example, taking a target farmland designated for monitoring by a planting entity in a certain region after user authorization as an example, the starting node corresponding to the root cause semantic event "soil mild drought" is first retrieved and located in the constructed causal event chain. The semantic event type corresponding to the inducing factor is obtained as soil moisture anomaly numerical semantic event. At the same time, the agronomic condition data of the target farmland authorized by the user is retrieved. Combining the agronomic impact dimension of the soil moisture anomaly event with the basic production factors of the target farmland, such as sandy loam soil type, drip irrigation method, and basal fertilizer application system, an agronomic attribution screening framework including soil moisture parameters, crop water stress response parameters, and nutrient availability parameters is constructed.

[0180] Then, starting with the mild soil drought event node, the four direct successor nodes were visited sequentially along the causal event chain: insufficient available nitrogen content in the soil, yellowing of crop leaf edges, wilting of crop leaves, and reduction in the number of effective ears per crop plant. For each direct successor node, its collection timestamp and location coordinates were extracted, as well as the collection timestamp and location coordinates of its upstream node, to obtain the time interval between adjacent nodes. The spatial distance between adjacent nodes was the same grid node. The results were compared with the agronomic transmission time range and physical field influence range of the corresponding event type to verify that the spatiotemporal continuity between all adjacent nodes conformed to agronomic laws.

[0181] Next, the semantic event types corresponding to each access node are extracted sequentially and matched dimension-by-dimensionally with the agronomic attribution screening framework to determine the categories of influencing factors that need to be screened for each node. User-authorized agronomic mechanism data from the target farmland is retrieved, and combined with the causal transmission relationship between each node and its upstream nodes, relevant influencing factors such as soil volumetric water content, soil available nitrogen content, leaf chlorophyll content, leaf water potential, and number of effective panicles per plant are screened from the multi-source semantic event feature parameters corresponding to each node.

[0182] Finally, following the order of nodes in the causal event chain, all relevant influencing factors of each node are collected sequentially and combined in an orderly manner according to the logical order of event transmission. This forms a complete set of structured information, including the initial soil volumetric water content parameters of the mild soil drought event, the nutrient availability parameters of the soil available nitrogen deficiency event, the chlorophyll content parameters of the crop leaf edge yellowing event, the leaf water potential parameters of the crop leaf wilting event, and the yield impact parameters of the crop single plant effective ear reduction event. This is the agronomic attribution information of the mild soil drought event.

[0183] like Figure 2 The diagram shown is a functional block diagram of an agricultural information real-time analysis system based on multi-source sensor data provided in an embodiment of the present invention.

[0184] The agricultural information real-time analysis system 100 based on multi-source sensor data described in this invention can be installed in an electronic device. Depending on the functions implemented, the agricultural information real-time analysis system 100 based on multi-source sensor data may include a data acquisition module 101, a semantic parsing module 102, an event parsing module 103, an event fusion module 104, a cause tracing module 105, and a report generation module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0185] In this embodiment, the functions of each module / unit are as follows: Data acquisition module 101: Collects environmental data, soil data and crop image data of the target farmland in real time through multi-source sensors, and encapsulates the environmental data and soil data into numerical data, and treats the crop image data as visual data; Semantic parsing module 102: Based on the agronomic conditions of the target farmland, it parses the numerical semantic events in the numerical data, identifies the phenotypic semantic events in the visual data, and merges the numerical semantic events and the phenotypic semantic events into multi-source semantic events; Event parsing module 103: Analyzes the temporal and positional relationships of the multi-source semantic events, divides the multi-source semantic events at the same position into temporal leading events and temporal following events, and constructs causal event pairs of the multi-source semantic events based on the temporal leading events and the temporal following events; Event fusion module 104: Connects the causal event pairs with shared event nodes to form a causal event chain of the target farmland, fuses the semantic event types and their transmission processes of each node in the causal event chain, and generates the contextual features of the target farmland; Cause tracing module 105: Based on the transmission path of each semantic event type in the context feature, traces the node dependency relationship of the causal event chain, locates the root semantic event of the current context feature according to the node dependency relationship, and analyzes the inducing factors of the context feature; Report generation module 106: Based on the root semantic event and the triggering factor, extract the agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain, and generate an agricultural information analysis report of the target farmland by combining the contextual features and the agronomic attribution information.

[0186] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0187] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0188] Furthermore, the functional modules in the various embodiments of the present invention 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 in the form of hardware plus software functional modules.

[0189] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0190] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for real-time analysis of agricultural information based on multi-source sensor data, characterized in that, The method includes: S1: Real-time acquisition of environmental data, soil data, and crop image data of the target farmland through multi-source sensors, and encapsulation of the environmental data and soil data into numerical data, and crop image data into visual data; S2: Based on the agronomic conditions of the target farmland, analyze the numerical semantic events in the numerical data, identify the phenotypic semantic events in the visual data, and merge the numerical semantic events and the phenotypic semantic events into multi-source semantic events; S3: Analyze the temporal and positional relationships of the multi-source semantic events, divide the multi-source semantic events at the same position into temporal leading events and temporal following events, and construct causal event pairs of the multi-source semantic events based on the temporal leading events and the temporal following events; S4: Connect the causal event pairs with shared event nodes to form a causal event chain of the target farmland, and fuse the semantic event types and transmission processes of each node in the causal event chain to generate the contextual features of the target farmland; S5: Based on the transmission path of each semantic event type in the context feature, trace the node dependency relationship of the causal event chain, locate the root semantic event of the current context feature according to the node dependency relationship, and analyze the inducing factors of the context feature. S6: Based on the root semantic event and the triggering factor, extract the agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain, and generate an agricultural information analysis report of the target farmland by combining the contextual features and the agronomic attribution information.

2. The method for real-time analysis of agricultural information based on multi-source sensor data as described in claim 1, characterized in that, The process of analyzing numerical semantic events in the numerical data and identifying phenotypic semantic events in the visual data based on the agronomic conditions of the target farmland includes: By combining the agronomic conditions of the target farmland with the growth parameters of the crops in the target farmland, the numerical data is verified, and outlier data points are screened out from the numerical data. Based on the abnormal data points and the state of the target farmland, the numerical semantic events of the target farmland are obtained; Key phenotypic features are extracted from the visual data, compared with the normal growth state of crops in the target farmland, and feature variation points of the visual data are obtained. These feature variation points are then summarized as phenotypic semantic events.

3. The method for real-time analysis of agricultural information based on multi-source sensor data as described in claim 1, characterized in that, The analysis of the temporal and positional relationships of the multi-source semantic events, dividing multi-source semantic events at the same position into temporal leading events and temporal successor events, includes: Extract the location coordinates of the occurrence of the multi-source semantic events, and aggregate the multi-source semantic events with the same location coordinates into a spatially consistent event set. Sort the semantic events in the spatially consistent event set according to their occurrence time to obtain a semantic event sequence. The semantic topic dependencies of adjacent events in the semantic event sequence are evaluated to obtain the dependency evaluation results of the semantic event sequence. Based on the dependency evaluation results, the events in the semantic event sequence are divided to obtain the temporal leading events and temporal successor events.

4. The method for real-time analysis of agricultural information based on multi-source sensor data as described in claim 1, characterized in that, The construction of causal event pairs for the multi-source semantic events based on the temporal leading event and the temporal successor event includes: Based on the agronomic mechanism of the target farmland, the transmission medium type of the temporal leading event and the temporal successor event in the same spatial location is verified to obtain the candidate causal events of the multi-source semantic event; By comparing the start time of the leading event with the start time of the subsequent events in the candidate causal events, and analyzing the time difference between the two and the rationality of the physical transmission rate of the transmission medium based on the comparison results, the temporally reasonable event pairs of the multi-source semantic events are obtained. The continuity of action and the directionality of response between the physical field influence boundary of the preceding event and the spatial region boundary gradient direction of the subsequent event in the temporally reasonable event pair are evaluated to obtain the causal event pair of the multi-source semantic event.

5. The method for real-time analysis of agricultural information based on multi-source sensor data as described in claim 1, characterized in that, The causal event pairs that share a common event node are linked together to form a causal event chain for the target farmland, including: The event nodes are sorted according to the chronological order of the events in the causal event pair to obtain the time node sequence of the causal event pair. Event nodes with empty inbound edge lists in the time node sequence are extracted as root candidate nodes. Starting from each of the root candidate nodes, trace subsequent events along the outgoing edge direction to obtain an initial causal chain segment. Locate the successor node with multiple incoming edge branches in the initial causal chain segment to obtain the branch convergence node of the initial causal chain segment. The agronomic contribution of each predecessor branch corresponding to the branch convergence node is compared to obtain the dominant branch and non-dominant branch corresponding to the branch convergence node. The non-dominant branch is merged into the link where the dominant branch is located to obtain a branchless merged causal chain segment. By connecting the spatiotemporal continuity between the end events and the beginning events of adjacent segments in the branchless merged causal chain segment, the causal event chain of the target farmland is obtained.

6. The method for real-time analysis of agricultural information based on multi-source sensor data as described in claim 1, characterized in that, The process of fusing the semantic event types and their transmission processes of each node in the causal event chain to generate contextual features of the target farmland includes: Extract the propagation characteristics of semantic event types of each node in the causal event chain, and filter the spatial density of the transmission process in the causal event chain to obtain the key transmission path of the transmission process; By integrating the key transmission paths and propagation characteristics of each node in the causal event chain, the contextual features of the target farmland are generated.

7. The method for real-time analysis of agricultural information based on multi-source sensor data as described in claim 1, characterized in that, The step of tracing the node dependencies of the causal event chain based on the transmission path of each semantic event type in the context features includes: Traverse the transmission paths in the context features in reverse order, and anchor the starting node and key interruption point of the transmission path; Based on the starting node, the continuity of semantic event types in the transmission path is analyzed one by one to obtain the dependency relationship between nodes in the transmission path and generate the node dependency relationship of the causal event chain.

8. The method for real-time analysis of agricultural information based on multi-source sensor data as described in claim 1, characterized in that, The step of locating the root semantic event of the current context feature based on the node dependency relationship and analyzing the triggering factors of the context feature includes: Analyze the dependency depth of each semantic event node in the node dependency relationship, and sort the semantic event nodes according to the dependency depth to obtain the event node sequence; The agronomic dominance of node event types in the event node sequence is analyzed, and combined with the spatial distribution aggregation degree assessment of the target farmland, the node with the largest dependency depth and the highest matching degree between the node event type and the agronomic conditions of the target farmland is identified as the root semantic event. Extract the region where the feature parameters of the root semantic event exceed the target crop growth standard parameters, and use the features of the region where the difference exceeds the target crop growth standard parameters as the inducing factors of the context features.

9. The method for real-time analysis of agricultural information based on multi-source sensor data as described in claim 1, characterized in that, The step of extracting agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain based on the root semantic event and the triggering factor includes: Locate the starting node of the root semantic event in the causal event chain, and determine the screening framework for agronomic attribution in the target farmland based on the semantic event type of the inducing factor and the agronomic conditions of the target farmland. Starting from the starting node, traverse along the propagation direction of the causal event chain to each direct successor node, and verify the spatiotemporal continuity between the direct successor node and the upstream node; Match the semantic event type with the filtering framework, and combine the agronomic mechanism of the target farmland to filter relevant influencing factors; The filtering results of all nodes in the combined sequence form the agronomic attribution information of the root semantic event.

10. A real-time agricultural information analysis system based on multi-source sensor data, characterized in that, The system includes: Data acquisition module: Collects environmental data, soil data and crop image data of the target farmland in real time through multi-source sensors, and encapsulates the environmental data and soil data into numerical data, and treats the crop image data as visual data; Semantic parsing module: Based on the agronomic conditions of the target farmland, it parses the numerical semantic events in the numerical data, identifies the phenotypic semantic events in the visual data, and merges the numerical semantic events and the phenotypic semantic events into multi-source semantic events; Event parsing module: Analyzes the temporal and positional relationships of the multi-source semantic events, divides the multi-source semantic events at the same position into temporal leading events and temporal following events, and constructs causal event pairs of the multi-source semantic events based on the temporal leading events and the temporal following events; Event fusion module: Connects the causal event pairs with shared event nodes to form a causal event chain of the target farmland, fuses the semantic event types and their transmission processes of each node in the causal event chain, and generates the contextual features of the target farmland; Cause tracing module: Based on the transmission path of each semantic event type in the context feature, trace the node dependency relationship of the causal event chain, locate the root semantic event of the current context feature according to the node dependency relationship, and analyze the triggering factors of the context feature; Report generation module: Based on the root semantic event and the triggering factor, extract the agronomic attribution information of the root semantic event node by node along the transmission direction of the causal event chain, and generate an agricultural information analysis report of the target farmland by integrating the contextual features and the agronomic attribution information.

Images