Intelligent agricultural production data processing method based on artificial intelligence

Abstract

The invention discloses an intelligent agricultural production data processing method based on artificial intelligence, and relates to the technical field of intelligent agriculture, and the method comprises the following steps: carrying out the construction of a standard feature vector set based on agricultural production data, and obtaining the standard feature vector set; performing farmland area abnormal state detection according to the standard feature vector set to obtain an abnormal detection result; performing clustering analysis according to the anomaly detection result to obtain an anomaly region division result; performing anomaly initiation factor analysis according to an anomaly region division result to obtain a causal variable set; according to the causal variable set, performing abnormal cause reasoning to obtain a causal reasoning result; and screening key factors according to a causal reasoning result to obtain a farmland key factor set. According to the method, the physical rationality and the dynamic adaptability of path reasoning are enhanced, the adaptability and the practicability of the model are improved, and the accuracy of reasoning of the abnormality cause is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of smart agriculture, and specifically to a method for processing smart agriculture production data based on artificial intelligence. Background Art

[0002] With the continuous advancement of precision agriculture and smart agriculture, a large amount of environmental, crop, physiological, and biochemical data from agricultural production processes is being collected in real time through multi-source sensors, remote sensing satellites, and agricultural IoT devices. This heterogeneous, multi-scale, and high-dimensional agricultural production data provides a foundation for intelligent agricultural management and data-driven decision support. However, the complex nonlinear dynamics and spatiotemporal coupling of agricultural production systems make it difficult to directly infer the causes of abnormal conditions such as pests and diseases, water stress, and soil degradation from the data.

[0003] In the existing technology, there are deficiencies in the reasoning of the causes of abnormal states: existing methods for reasoning about the causes of abnormal states mostly use graphical models based on structural learning, such as PC algorithms, GES, LiNGAM, etc., or use Granger causal models based on time series analysis. They lack the characterization of dynamic indicators such as disturbance propagation intensity and path influence in causal path reasoning, and cannot reflect how variable disturbances are actually transmitted to abnormal states. They cannot adapt to the highly dynamic and regional heterogeneous characteristics of agricultural systems. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an artificial intelligence-based smart agricultural production data processing method to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides a method for processing smart agricultural production data based on artificial intelligence, comprising the following steps: S1. Construct a standard feature vector set based on agricultural production data to obtain a standard feature vector set; S2. Detect abnormal conditions in the farmland area based on the standard feature vector set to obtain abnormality detection results; S3. Perform cluster analysis based on the anomaly detection results to obtain the abnormal area division results; S4. Analyze the factors causing abnormalities based on the abnormal area division results to obtain a set of causal variables; S5. Infer the cause of the anomaly based on the causal variable set to obtain the causal reasoning result; S6. Screen key factors based on the causal reasoning results to obtain a set of key factors for farmland.

[0006] To further optimize this technical solution, the abnormal state detection of the farmland area in step S2 includes: Based on the obtained standard feature vector set, abnormal state detection is performed on the farmland area. First, spatial representation transformation is performed to characterize the spatial abnormal performance pattern in the agricultural production scene. Then dynamic time gating is performed to adjust the current response of the model to historical influences. Then, spatiotemporal fusion representation generation is performed to integrate the spatial propagation characteristics and temporal dynamic response at the current moment. Finally, classification output is performed to judge the abnormal prediction probability of each farmland grid unit and obtain the anomaly detection result.

[0007] To further optimize this technical solution, the spatial representation transformation includes: ; in: : spatial adjacency matrix; : The standard eigenvector matrix at time t, the matrix dimension is M×N, N is the number of feature dimensions of each farmland grid unit; : spatial weight transformation matrix; : spatial embedding representation matrix at time t; : activation function.

[0008] Further optimizing this technical solution, the dynamic time gating includes: ; in: : global dynamic time gating vector at time t; : time dynamic weight matrix; : bias vector; : Find the average function; : Sigmoid activation function.

[0009] Further optimizing this technical solution, the generation of the spatiotemporal fusion representation includes: ; in: : The fused spatiotemporal feature representation matrix; : M-dimensional all-one vector; : fusion weight hyperparameters; : Replication of the temporal gating vector.

[0010] Further optimizing this technical solution, the classification output includes: ; in: : Classification result output of farmland grid unit at time t; : classification weight matrix; : Perform Softmax transformation on the output vector of each farmland grid cell.

[0011] To further optimize this technical solution, the analysis of the abnormality-causing factors in S4 includes: Based on the obtained abnormal area division results, artificial perturbations are introduced into the agricultural characteristic variables in the abnormal areas to simulate slight fluctuations of the agricultural characteristic variables. The causal impact scoring model is used to analyze the abnormality triggering factors, and finally the abnormality triggering factors with the most causal impact are screened out to form the causal variable set for each area.

[0012] To further optimize this technical solution, the causal impact scoring model includes: ; in: : The causal score of the nth feature of the jth abnormal region; : The probability of abnormality before the introduction of artificial perturbation in the jth abnormal area; : The abnormal probability after the j-th abnormal area introduces artificial perturbation.

[0013] To further optimize this technical solution, the abnormal probability includes: ; ; in: : disturbance intensity; : The nth characteristic variable of the jth abnormal area; : time step of the time series; : the jth abnormal area; : The disturbance characteristic data matrix of the j-th abnormal area; : The original feature data matrix of the abnormal area of ​​the j-th abnormal area; : Abnormal probability function.

[0014] To further optimize this technical solution, the disturbance characteristic data matrix includes: ; in: : The n-th dimension feature variable vector.

[0015] In a second aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of a method for processing smart agricultural production data based on artificial intelligence as described in the first aspect of the present invention are implemented.

[0016] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of a method for processing smart agricultural production data based on artificial intelligence as described in the first aspect of the present invention are implemented.

[0017] Compared with the existing technology, the present invention provides a method for processing smart agricultural production data based on artificial intelligence, which has the following beneficial effects: This AI-based smart agricultural production data processing method uses a dynamic disturbance transmission causal chain model, machine learning and deep learning to simulate the system response after causal variable disturbance, forming a response chain between variables in time and space, enhancing the physical rationality and dynamic adaptability of path reasoning. Each plot is modeled separately to match geographical heterogeneity, which is more suitable for field-level management, improving the adaptability and practicality of the model, and modeling the variable disturbance transmission to improve the accuracy of reasoning about the cause of anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of a method for processing smart agricultural production data based on artificial intelligence proposed by the present invention; Figure 2 This is a flow chart of abnormal state detection in an artificial intelligence-based smart agricultural production data processing method proposed by the present invention; Figure 3 This is a flow chart of a causal impact scoring model for an artificial intelligence-based smart agricultural production data processing method proposed in the present invention; Figure 4 This is a flow chart of a dynamic disturbance transmission causal chain model of an artificial intelligence-based smart agricultural production data processing method proposed in the present invention. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it constitute a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1: Reference Figures 1 to 4 , which is the first embodiment of the present invention, provides a method for processing smart agricultural production data based on artificial intelligence, comprising the following steps: S1. Construct a standard feature vector set based on agricultural production data to obtain a standard feature vector set.

[0024] In this embodiment, the construction of the standard feature vector set includes: In agricultural production data processing, raw data from different sources often have different formats, scales, time intervals, and spatial distributions. This data heterogeneity complicates subsequent data analysis and modeling. To address this issue, the core purpose of this step is to convert heterogeneous data sources into a unified set of standard feature vectors to facilitate data analysis and modeling.

[0025] The goal of this step is to convert raw agricultural production data (such as meteorological data, crop growth data, and soil data) into a standardized input format. Raw agricultural production data typically comes from multiple different sources, including multispectral imagery (image data used to monitor crop growth status, often stored in pixel form, containing image data in bands such as red, green, blue, and near-infrared light), meteorological data (including real-time environmental information such as temperature, humidity, wind speed, and precipitation, usually continuous time series data), soil data (including soil conductivity, humidity, pH value, water content, and nutrient element (nitrogen, phosphorus, potassium, etc.) concentrations), and crop growth data (such as chlorophyll index, crop growth, and aboveground and underground growth).

[0026] Different data sources have different frequencies and formats for temporal collection. By selecting a unified time scale (such as hourly or daily), the timestamps of all data are unified so that they can be aligned on the same time axis. For those cases where the data intervals are different, linear interpolation or spline interpolation (such as Kriging interpolation) is used to fill in the missing data points so that all data are complete and consistent on the unified time axis. Different agricultural production data usually have different units and dimensions (such as soil moisture in percentage and temperature in degrees Celsius). To eliminate these differences, Z-score normalization (removing the mean and dividing by the standard deviation) or Min-Max normalization (scaling data values ​​to between 0 and 1) is used to perform feature standardization, thereby compressing the range of all data to the same scale to prevent certain features from having too large an impact on the model due to excessive dimensions. Different agricultural production data often have spatial distribution characteristics. For example, data such as soil moisture and crop growth are typically collected based on geographic coordinates. For unified analysis, the data needs to be mapped to a unified spatial grid. Farmland is typically divided into grid cells (e.g., 10m×10m grid cells), and the data within each grid cell is merged using interpolation methods. For multispectral image data, pixel data is spatially aligned according to geographic coordinates, and each pixel in the image is mapped to the farmland grid. For other data types (such as meteorological data and soil data), Kriging interpolation or nearest neighbor interpolation is often used to fill in grid cells with no data. Missing data is inevitable in agricultural production. Incomplete data may result from instrument failure, weather interference, or human error. To ensure that models can handle this incomplete data, effective missing value handling methods are required. Interpolation methods are often used to fill missing values, such as linear interpolation or Kriging interpolation for missing meteorological or soil data. These methods can infer the values ​​of the missing parts based on the trends of the known data, thereby preventing the impact of missing values ​​on model training.

[0027] After synchronizing and standardizing the raw agricultural production data, all data is converted into a unified feature vector. The feature vector of a farmland grid cell may include meteorological data, soil data, crop growth data, multispectral image data, and other data. Converting this multidimensional data into a standardized feature vector ultimately yields a standardized feature set for each grid cell. This set includes standardized data on features from different time steps or data sources.

[0028] S2. Detect abnormal conditions in the farmland area based on the standard feature vector set to obtain abnormality detection results.

[0029] In this embodiment, the abnormal state detection of the farmland area includes: The standard feature vector set obtained in step S1 has been standardized and spatiotemporally gridded for multi-source heterogeneous data, including meteorological data, crop growth status, soil indicators, and multispectral imagery. The occurrence of abnormal conditions, such as crop diseases, soil drought, and flooding, is often influenced by both spatial neighborhood influences (e.g., contagion and spread from neighboring plots) and temporal dynamics (e.g., continuous rainfall or high temperatures). This step uses the obtained standard feature vector set to detect abnormal conditions in farmland areas. First, a spatial representation transformation is performed to characterize spatial abnormality patterns in agricultural production scenarios, such as disease transmission paths and soil anomalies. Dynamic temporal gating is then performed to adjust the model's current response to historical influences and, to a certain extent, mitigate interference from redundant temporal features. A spatiotemporal fusion representation is then generated to integrate the current spatial propagation characteristics with the temporal dynamic response. Finally, a classification output is generated to determine the anomaly prediction probability for each farmland grid cell, resulting in anomaly detection results.

[0030] Furthermore, the spatial representation transformation includes: ; in: : Spatial adjacency matrix, which represents the topological connection relationship between any two farmland grid cells. A value of 1 indicates that there is spatial adjacency, and a farmland grid cell is not spatially adjacent to itself. A value of 0 is obtained from the geographical adjacency relationship between farmland grid cells. The matrix dimension is M×M, where M is the number of farmland grid cells. : The standard eigenvector matrix at time t, the matrix dimension is M×N, N is the number of eigenvector dimensions of each farmland grid unit; : Spatial weight conversion matrix, used to map the feature vectors of farmland grid cells into an H-dimensional spatial representation. The size of the dimension H is set by yourself. The purpose is to improve the model's ability to express complex spatiotemporal features, provide sufficient dimensions to support nonlinear learning and control model capacity, and avoid insufficient expressiveness due to too low dimensions or overfitting due to too high dimensions. It is usually set in the range of 8 to 128 dimensions, and the matrix dimension is N × H; : The spatial embedding representation matrix at time t represents the spatial representation of each farmland grid unit after neighborhood propagation and feature compression. The matrix dimension is M×H. The matrix integrates the features of itself and multiple spatially adjacent farmland grid units, realizing information fusion in the spatial dimension. Therefore, when the features of the surrounding farmland grid units change due to abnormalities, even if the current farmland grid unit does not have an abnormality, the abnormalities of the surrounding farmland grid units can be discovered through the matrix, and the abnormal risks can be predicted in advance to achieve early abnormality perception. For example, when the features of the surrounding farmland grid units of a certain farmland grid unit change, the spatial embedding representation matrix of the farmland grid unit will also change. : Activation function, representing max(0,x), retains the original input in the positive interval and outputs 0 in the negative interval, which is used to increase nonlinear capabilities.

[0031] Furthermore, the dynamic time gating includes: Establish a global time gating mechanism to extract the overall trend of all spatial units at the current moment, thereby controlling the response of subsequent models to the current features; ; in: : The global dynamic time gating vector at time t, which represents the dynamic state of the entire agricultural area at time t, with a vector dimension of H; : The time-dynamic weight matrix is ​​used to map the N-dimensional average feature vector to the H-dimensional gated representation, with a matrix dimension of H×N; : bias vector, the vector dimension is H; : Averaging function, used to average the characteristic vectors of all farmland grid cells at time t to reflect the overall state characteristics. , represents the N-dimensional feature vector of the i-th farmland grid unit at time t. The N-dimensional average feature vector is obtained by averaging the feature vectors of all farmland grid units dimension by dimension. : Sigmoid activation function, used to map the gate value to [0, 1].

[0032] Furthermore, the generation of the spatiotemporal fusion representation includes: ; in: : The fused spatiotemporal features represent the matrix, and the matrix dimension is M×H; : M-dimensional all-1 vector, used for copy operation; : Fusion weight hyperparameter, which can be set to a fixed value to adjust the importance of temporal and spatial information; : The replication of the time-gated vector means replicating the time-gated vector with a dimension of 1×H M times so that it is fused with each farmland grid unit.

[0033] Furthermore, the classification output includes: ; in: : The classification result output of the farmland grid unit at time t, that is, the probability of each classification category. The matrix dimension is M×C, where C is the number of categories (for example, if the classification categories are normal and abnormal, then C is 2). The i-th row is the classification result vector of the i-th farmland grid unit at time t; : classification weight matrix, the matrix dimension is H×C; : Perform Softmax transformation on the output vector of each farmland grid cell so that it satisfies the probability distribution, that is, the sum of the elements in each row is 1.

[0034] The model describes how to detect abnormal conditions in farmland areas based on a set of standard feature vectors.

[0035] Traditional anomaly detection methods usually ignore spatial and temporal dependencies and often only consider data characteristics of a single dimension. Agricultural production data has significant spatiotemporal dependencies. Characteristics such as soil moisture and crop growth change over time and have strong local correlations in space. Therefore, traditional anomaly detection methods may lead to insufficient detection accuracy. This model can simultaneously capture the spatial dependency and temporal variation characteristics of multidimensional data in agricultural production, thereby significantly improving the accuracy and real-time performance of anomaly detection.

[0036] The steps for using this model include: Data acquisition: obtaining standardized and spatiotemporal gridded data based on the standard feature vectors obtained in step S1; Spatiotemporal feature fusion: Calculate the spatial representation of the farmland area based on the acquired data and time gate vector , and fuse them to obtain the spatiotemporal feature representation matrix ; Anomaly detection: Based on the calculated spatiotemporal features, anomaly detection is performed on each farmland grid unit to obtain the classification results of each farmland grid unit. For example, if the classification categories are normal and abnormal, the probability of normal and abnormality of each farmland grid unit is expressed as follows: , indicating that the probability that the 23rd farmland grid unit is judged to be in an abnormal state at time t is 85%, and the probability that it is in a normal state is 15%. An abnormal threshold is set, and the obtained abnormal probability is compared with the abnormal threshold. If the abnormal probability is greater than the threshold, the farmland grid unit is judged to be abnormal.

[0037] S3. Perform cluster analysis based on the anomaly detection results to obtain the abnormal area division results.

[0038] In this embodiment, cluster analysis includes: In step S2, it is determined whether each farmland grid unit is in an abnormal state, but the specific abnormality generated by each farmland grid unit is still unclear. Therefore, the goal of this step is to divide the farmland grid with unknown abnormal state into non-overlapping areas in space according to the abnormality type, and clarify the distribution boundaries of various types of problems, providing a spatial semantic basis for subsequent problem cause reasoning.

[0039] This step uses the density-based spatial clustering of applications with noise (DBSCAN) algorithm to perform spatial cluster analysis and determine the outlier regions. This algorithm requires no pre-defined number of clusters, can handle spatial noise points (i.e., identify "isolated outliers"), and offers flexible clustering. It's suitable for complex field shapes (non-rectangular, with irregular boundaries, etc.) found in real farmland, and can naturally form spatially continuous outlier regions.

[0040] According to the anomaly detection results obtained in step S2, an abnormal unit set is extracted from it, and the abnormal units are mapped to their two-dimensional spatial coordinates (the spatial coordinates can be derived from the center point of the plot, remote sensing pixel coordinates, longitude and latitude, or plane coordinate system) to form a coordinate set. Then, based on the actual field size and remote sensing image resolution, density clustering parameters are set, including the neighborhood radius (controlling the spatial perception range) and the minimum number of neighborhood points (used to control the clustering threshold). Then, a clustering operation is performed, the abnormal unit set is input, and the density clustering algorithm is run to obtain the clustering result. The clustering result is multiple non-overlapping abnormal areas. Finally, the boundary of each abnormal area is calculated, and the boundary envelope is extracted. The α-Shape or Convex Hull method (for a small number of points) can be used, or the contour line method (for a dense number of points) can be used to obtain the boundary contour line of each abnormal area. According to the abnormal area data and boundary contour line data obtained by cluster analysis, a complete abnormal area division result is finally obtained.

[0041] S4. Analyze the factors causing the abnormality based on the abnormal area division results to obtain a set of causal variables.

[0042] In this embodiment, the analysis of abnormality-causing factors includes: Based on the obtained abnormal area division results, artificial perturbations are introduced to the agricultural characteristics in the abnormal areas to simulate slight fluctuations of agricultural characteristics. The causal impact scoring model is used to quantify its impact on the abnormal judgment output, and an abnormality triggering factor analysis is performed. Finally, the abnormality triggering factors with the most causal impact are screened out to form the causal variable set for each area.

[0043] Furthermore, the causal impact scoring model includes: ; in: : The causal score of the nth feature of the jth abnormal region. The larger the value, the stronger the causal relationship between the feature vector and the abnormality. : The abnormal probability of the jth abnormal region before the introduction of artificial perturbation, obtained by the regional abnormal probability calculated in step S2; : The abnormal probability after the j-th abnormal area introduces artificial perturbation.

[0044] Furthermore, the abnormal probability includes: ; ; in: : disturbance intensity, which indicates the intensity of the introduced artificial perturbation; : The nth feature of the jth abnormal region; : time step of the time series; : the jth abnormal area; : The disturbance feature matrix of the jth abnormal region represents the feature vector matrix constructed after the introduction of artificial perturbation. The matrix dimension is T×N, where T is the length of the time series data. The time series should be selected to cover the change process from no abnormality to the presence of the abnormality. N is the number of feature dimensions of the abnormal region. : The original feature matrix of the abnormal region of the j-th abnormal region, that is, the feature vector matrix constructed before the introduction of artificial perturbation. This matrix includes the feature vectors of N dimensions of the abnormal region at T time steps, and the matrix dimension is T×N; : Abnormal probability function, which indicates the probability of abnormality in the region according to the classification result obtained in step S2. Since the abnormal region is composed of multiple farmland grid units, the original feature matrix of the abnormal region is and the perturbation characteristic matrix All of them can be converted into M×N dimensional standard feature vector matrices under multiple time steps, so that the abnormality probability can be calculated using the model in step S2.

[0045] Furthermore, the disturbance characteristic data matrix includes: ; in: : The n-th dimension feature vector, obtained through the regional standard feature vector set obtained in step S1.

[0046] This model describes how to analyze the impact of characteristic variables on anomaly induction by perturbing them.

[0047] Traditional methods for analyzing the factors causing anomalies are usually only applicable to large sample scenarios with stable variables and stable structural relationships. They are insensitive to complex nonlinear relationships, have difficulty capturing the causal characteristics of local areas, have low judgment accuracy in environments with high noise and missing data, and the output results are mostly abstract structural diagrams or statistical indicators that are difficult to understand directly. However, this model supports complex nonlinear relationships between multiple variables in agricultural production data and can be used for agricultural data with fewer samples and larger variable noise. It independently evaluates causal variables for each abnormal area, improves adaptability and judgment accuracy, and can obtain scores for each causal variable, which facilitates intuitive interpretation and understanding and improves practicality.

[0048] The steps for using the above model include: Disturbance construction: Extract the characteristic variable matrix based on the abnormal area obtained in step S3 , and inject perturbations into the characteristic variables , to perturb the characteristic matrix structure; Abnormal probability calculation: Based on the constructed disturbance characteristic matrix, use the model in step S2 to calculate the abnormal probability after the perturbation is injected ; Causal impact score: the probability of anomaly after injecting perturbations and the original abnormal probability Compare and calculate to get causal score The larger the value of the causal score, the stronger the causal influence, indicating that the causal relationship between the characteristic variable and the anomaly is stronger, and it is more likely to be the key characteristic variable factor causing the anomaly. A causal score threshold is set, and the characteristic variables with a causal score greater than this threshold are combined to form a causal variable set. For example, if there is an anomaly in a certain block, and the causal variables obtained are soil moisture, vegetation index NDVI, and pest and disease index, then the anomaly of this block may be related to the continuous decline in soil moisture, the decline in vegetation index NDVI, and the sharp increase in pest and disease index.

[0049] S5. Infer the cause of the anomaly based on the causal variable set to obtain the causal reasoning result.

[0050] In this embodiment, the reasoning of the cause of the exception includes: The causal variable set obtained in step S4 clarifies the variable factors that trigger the anomaly. In this step, based on the causal variable set, a dynamic disturbance transfer causal chain model is used to infer the cause of the anomaly. This analyzes and clarifies how the variables in the causal variable set trigger the anomaly, and obtains the causal inference results for each region. The causal inference results include the anomaly-causing causal chain and causal chain score for each region, which explain the reason why the variable triggers the anomaly. The causal chain includes the starting characteristic variable, intermediate characteristic variables, and key characteristic variables that lead to the abnormal state. The starting characteristic variable is the source of the causal path and does not directly cause the anomaly, but has an excitation effect on the intermediate characteristic variables. The intermediate characteristic variables do not directly cause the anomaly, but play an intermediate amplification, transmission, or synergistic role. The key characteristic variable is close to the abnormal result itself, or its fluctuation directly triggers the anomaly. The key characteristic variable that causes the anomaly is obtained through the causal score of the abnormal region characteristics. Combined with the disturbance transfer strength between different characteristic variables, the intermediate characteristic variables between the starting variable and the key characteristic variable are inferred step by step until the starting characteristic variable that causes the abnormal state is found, thus forming a complete causal chain.

[0051] Furthermore, the dynamic disturbance transmission causal chain model includes: ; in: : Causal chain score, used to determine the impact of the path on the abnormal state. The higher the score, the greater the impact. : A path chain from characteristic variables to abnormal nodes. For example, soil temperature changes affect soil moisture changes, which in turn affect pest and disease indexes and ultimately lead to abnormalities. : Disturbance transfer intensity, reflecting the influence of variable u on variable v; : The causal score of variable v, obtained by the model calculation in step S4.

[0052] Furthermore, the disturbance transmission intensity includes: ; in: : Disturbance transfer intensity, reflecting the influence of variable p on variable q; : the value of variable q at time t in region j; : Apply a disturbance to the variable p in region j The value of the variable q at the next t moments; : The length of time series data. For example, agricultural data is sampled once every hour, so T=24 for 24 hours.

[0053] This model describes how to infer the causes of anomalies by analyzing the influence between variables.

[0054] Traditional anomaly-causing reasoning usually uses methods based on time series prediction or structural equation modeling. These methods are difficult to characterize the complex coupling of agricultural multivariables, are not suitable for high-dimensional, nonlinear interference systems, and are not suitable for agricultural environments at the plot level with strong spatial heterogeneity. It is difficult to dynamically adjust and model the transmission process of disturbances. This model can realize the integration of multi-source and multi-modal data, and is suitable for mixed scenarios of continuous variables, discrete variables and remote sensing indices. Each plot is modeled separately to match geographical heterogeneity, which is more suitable for field-level management, improves the adaptability and practicality of the model, realizes the modeling of disturbance transmission, and improves the accuracy of anomaly-causing reasoning. By simulating the system response after the causal variable is disturbed, a real response chain between variables in time and space is formed, breaking through the problem that traditional static causal diagrams cannot reflect the dynamic influence path of variables, and enhancing the physical rationality and dynamic adaptability of path reasoning.

[0055] Uses of the model include: Data acquisition: Obtain the causal variable set from step S4 to obtain the variables that have an impact on the occurrence of the anomaly; Disturbance transfer calculation: Add disturbances to each variable independently to simulate the response changes of other variables. For example, if the temperature rises by 1°C, does the vegetation index (NDVI) fluctuate significantly? The disturbance transfer intensity between variables is calculated. The causal score is calculated based on the causal impact scoring model in step S4 to obtain the causal relationship between the variable and the anomaly. Based on the disturbance transfer, a causal path chain is constructed to reflect the path from the variable to the anomaly. For example, the vegetation index (NDVI) -> pest and disease index -> anomaly, or the temperature -> soil moisture -> vegetation index (NDVI) -> anomaly, etc. Anomaly cause reasoning: Based on the calculated disturbance transmission intensity and causal score, the causal chain score of each causal path chain is calculated. The higher the score, the greater the impact of the path on the abnormal state and the more it can reflect how the variable causes the anomaly. For example, the score of the causal chain of vegetation index NDVI -> pest and disease index -> anomaly is lower than that of temperature -> soil moisture -> vegetation index NDVI -> anomaly, indicating that the cause of the anomaly in this area may be temperature changes causing soil moisture changes, which in turn lead to changes in vegetation index NDVI, ultimately causing the anomaly.

[0056] S6. Screen key factors based on the causal reasoning results to obtain a set of key factors for farmland.

[0057] In this embodiment, the key factor screening includes: Based on the causal reasoning results obtained in step S5, a disturbance causal graph with direction and weight is obtained. The node set represents the causal variables, and the edge set represents the disturbance path. The PageRank algorithm is used to calculate the structural importance of each node in the graph. The higher the PageRank score of a variable, the more core its position in the entire causal propagation graph and the more likely it is to cause a series of abnormalities downstream. A scoring threshold is set to screen out nodes with scores greater than the threshold, that is, the screened out key factor variables, forming a set of key factors for farmland. The variables in this set are the factors that deserve the most attention in subsequent agricultural regulation, intervention, or risk assessment.

[0058] Example 2: This embodiment also provides a computer device, which is suitable for a method for processing smart agricultural production data based on artificial intelligence, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a method for processing smart agricultural production data based on artificial intelligence as proposed in the above embodiment.

[0059] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements an artificial intelligence-based smart agricultural production data processing method proposed in the above embodiment.

[0060] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0061] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0062] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0063] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0064] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0065] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for processing smart agricultural production data based on artificial intelligence, characterized in that: The following steps are involved: S1. Construct a standard feature vector set based on agricultural production data to obtain a standard feature vector set; S2. Detect abnormal conditions in the farmland area based on the standard feature vector set to obtain abnormality detection results; S3. Perform cluster analysis based on the anomaly detection results to obtain the abnormal area division results; S4. Analyze the factors causing abnormalities based on the abnormal area division results to obtain a set of causal variables; S5. Infer the cause of the anomaly based on the causal variable set to obtain the causal reasoning result; S6. Screen key factors based on the causal reasoning results to obtain a set of key factors for farmland.

2. The method for processing smart agricultural production data based on artificial intelligence according to claim 1, characterized in that: The abnormal state detection of the farmland area in step S2 includes: Based on the obtained standard feature vector set, abnormal state detection is performed on the farmland area. First, spatial representation transformation is performed to characterize the spatial abnormal performance pattern in the agricultural production scene. Then dynamic time gating is performed to adjust the current response of the model to historical influences. Then, spatiotemporal fusion representation generation is performed to integrate the spatial propagation characteristics and temporal dynamic response at the current moment. Finally, classification output is performed to judge the abnormal prediction probability of each farmland grid unit and obtain the anomaly detection result.

3. The method for processing smart agricultural production data based on artificial intelligence according to claim 2, characterized in that: The spatial representation transformation includes: ； in: : spatial adjacency matrix; : The standard eigenvector matrix at time t, the matrix dimension is M×N, N is the number of eigenvector dimensions of each farmland grid unit; : spatial weight transformation matrix; : spatial embedding representation matrix at time t; : activation function.

4. The method for processing agricultural production data based on artificial intelligence according to claim 2, characterized in that: The dynamic time gating includes: ； in: : global dynamic time gating vector at time t; : time dynamic weight matrix; : bias vector; : Find the average function; : Sigmoid activation function.

5. The method for processing agricultural production data based on artificial intelligence according to claim 2, characterized in that: The generation of the spatiotemporal fusion representation includes: ； in: : The fused spatiotemporal feature representation matrix; : M-dimensional all-one vector; : fusion weight hyperparameters; : Replication of the temporal gating vector.

6. The method for processing smart agricultural production data based on artificial intelligence according to claim 2, characterized in that: The classification output includes: ； in: : Classification result output of farmland grid unit at time t; : classification weight matrix; : Perform Softmax transformation on the output vector of each farmland grid cell.

7. The method for processing smart agricultural production data based on artificial intelligence according to claim 1, characterized in that: The analysis of the abnormality-causing factors in S4 includes: Based on the obtained abnormal area division results, artificial perturbations are introduced into the agricultural characteristic variables in the abnormal areas to simulate slight fluctuations of the agricultural characteristic variables. The causal impact scoring model is used to analyze the abnormality triggering factors, and finally the abnormality triggering factors with the most causal impact are screened out to form the causal variable set for each area.

8. The method for processing smart agricultural production data based on artificial intelligence according to claim 7, characterized in that: The causal impact scoring model includes: ； in: : The causal score of the nth feature of the jth abnormal region; : The probability of abnormality before the introduction of artificial perturbation in the jth abnormal area; : The abnormal probability after the j-th abnormal area introduces artificial perturbation.

9. The method for processing smart agricultural production data based on artificial intelligence according to claim 8, characterized in that: The abnormal probability includes: ； ； in: : disturbance intensity; : The nth feature of the jth abnormal region; : time step of the time series; : the jth abnormal area; : The disturbance characteristic data matrix of the j-th abnormal area; : The original feature data matrix of the abnormal area of ​​the j-th abnormal area; : Abnormal probability function.

10. The method for processing smart agricultural production data based on artificial intelligence according to claim 9, characterized in that: The disturbance characteristic data matrix includes: ； in: : The n-th dimension feature variable vector.

Images