Crop disease and pest diagnosis system based on knowledge graph

Abstract

The invention discloses a crop disease and pest diagnosis system based on a knowledge graph, and relates to the technical field of disease and pest diagnosis. A data acquisition module acquires description information, image data and environment data of crops; the data processing module is used for extracting symptom keywords, disease and pest expansion rate fluctuation characteristics and soil EC value stress characteristics; the knowledge graph construction module is used for establishing an association network comprising disease and pest symptom nodes, environmental condition nodes and dynamic feature nodes; the disease and pest distinguishing module evaluates the distinguishing accuracy of disease and pest similar symptoms in combination with dynamic characteristics and environmental characteristics; the diagnosis result division module is used for accurately dividing diagnosis results and taking corresponding treatment measures; and the optimization management module dynamically optimizes a knowledge graph and a classification model based on fixed time period analysis of an incomplete accuracy diagnosis result, so that the accuracy of disease and insect pest diagnosis is remarkably improved, prevention and treatment errors caused by misdiagnosis are avoided, and intelligent technical support is provided for agricultural production.

Description

Technical Field

[0001] The present invention relates to the technical field of pest and disease diagnosis, and particularly to a crop pest and disease diagnosis system based on a knowledge graph. Background Art

[0002] Crop pest and disease diagnosis based on a knowledge graph is a method that uses artificial intelligence technology and knowledge engineering to systematize and structure the knowledge system in the field of crop pests and diseases, construct a knowledge network that can be understood and reasoned by machines, and perform diagnosis and prediction through intelligent algorithms. A knowledge graph is a network graph structure that represents knowledge as nodes (entities) and edges (relationships). In crop pest and disease diagnosis, the construction process of the knowledge graph includes extracting knowledge related to pests and diseases from agricultural literature, expert knowledge, and actual data to form a complete semantic network. For example, "rice" as a node is connected to "rice blast" through the "susceptible pest and disease" relationship; "rice blast" is in turn connected to "grayish-brown spots appear on the leaves" through the "typical symptom" relationship. This structured knowledge graph can systematically integrate information such as the classification, symptoms, pathogenesis, inducing factors, environmental factors, and control measures of pests and diseases, providing rich background knowledge for diagnosis.

[0003] The diagnosis process based on a knowledge graph mainly relies on reasoning algorithms and intelligent analysis techniques. When farmers upload pictures of crops or describe pest and disease symptoms, the system will match the input information with the knowledge in the knowledge graph, and through logical reasoning or pattern recognition techniques, find the most likely pests and diseases and their causes. For example, when the system detects that the input description is "irregular grayish-white spots appear on rice leaves", it will combine meteorological data and the occurrence rules of pests and diseases, and through the association inference of the knowledge graph, obtain a diagnosis result that may be "rice blast" or "bacterial leaf blight". Subsequently, the system can also provide targeted control suggestions, such as recommending the use of appropriate pesticides or adjusting farming methods.

[0004] The prior art has the following deficiencies: The diagnosis of the knowledge graph mainly relies on the matching between symptom descriptions and known pest and disease knowledge. However, in actual agricultural scenarios, different pests and diseases may exhibit similar symptoms. For example, "irregular grayish-white spots appear on rice leaves" may be either rice blast, bacterial leaf blight, or a non-pest and disease phenomenon caused by environmental stress (such as saline-alkali stress, phytotoxicity, or nutrient deficiency). When the system lacks more detailed graph nodes (such as specific pest and disease identification indicators) or actual image data, it may lead to misdiagnosis, thus giving farmers incorrect control suggestions, such as using the wrong pesticides or measures, further exacerbating the damage to crops. Summary of the Invention

[0005] The object of the present invention is to provide a crop pest and disease diagnosis system based on a knowledge graph to solve the deficiencies in the background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: A crop pest and disease diagnosis system based on a knowledge graph, comprising a data acquisition module, a data processing module, a knowledge graph construction module, a pest and disease differentiation module, a diagnosis result classification module, and an optimization management module; The data acquisition module is used to obtain pest and disease description information, pest and disease image data, and environmental data of crops, and the environmental data includes meteorological data and soil data; The data processing module is used to perform natural language processing on the pest and disease description information to extract symptom keywords, process the pest and disease image data through computer vision technology to extract the fluctuation characteristics of the pest and disease expansion rate, and obtain the soil EC value stress characteristics after processing the environmental data; The knowledge graph construction module is used to establish a knowledge graph based on pest and disease knowledge, and the knowledge graph includes pest and disease symptom nodes, environmental condition nodes, dynamic feature nodes, and the association relationships between the nodes; The pest and disease differentiation module is used to match the extracted symptom keywords with the symptom information stored in the knowledge graph, differentiate the pests and diseases according to the matching results, and evaluate the differentiation accuracy of similar pest and disease symptoms in combination with the fluctuation characteristics of the pest and disease expansion rate and the soil EC value stress characteristics; The diagnosis result classification module divides the accuracy of the pest and disease diagnosis result according to the evaluation result, divides the pest and disease diagnosis result into accurate diagnosis results, incomplete accurate diagnosis results, and inaccurate diagnosis results, and performs corresponding processing; The optimization management module further analyzes the differentiation accuracy of similar pest and disease symptoms within a fixed time period for incomplete accurate diagnosis results, and dynamically optimizes the knowledge graph and classification model according to the analysis results to improve the accuracy of crop pest and disease diagnosis.

[0007] Preferably, in the data processing module, after analyzing the extracted fluctuation characteristics of the pest and disease expansion rate, a pest and disease expansion rate fluctuation index is generated, and the acquisition method of the pest and disease expansion rate fluctuation index is as follows: Obtain the disease expansion rate data in different time periods and construct two time series: is the reference expansion rate sequence, representing the standard disease expansion rate pattern; is the currently collected disease expansion rate sequence, and each rate point represents the change rate of the lesion area in a certain time period. For any two points , calculate the local distance: ; where is the point and The absolute distance; Construct a cost matrix D, where D is an m×n matrix, and each element D(i,j) represents the minimum cumulative distance of the path from the starting point (1,1) to the point (i,j). The initial cost is: , ; , ; Fill the matrix sequentially through dynamic programming, and cumulatively calculate the minimum matching cost between each point; Define the optimal path as a sequence of point pairs, and the conditions are: represents the starting point; represents the ending point. The path is continuous and cannot jump. The total cost of the path is: ; Calculate the pest and disease spread rate fluctuation index, and the expression is: ; In the formula, is the path length, and WDC is the pest and disease spread rate fluctuation index.

[0008] Preferably, in the data processing module, after analyzing the extracted soil EC value stress characteristics, a soil EC value stress anomaly index is generated. The method for obtaining the soil EC value stress anomaly index is: Collect the sampling data set of soil EC values: input data , each represents the soil EC value of a sampling point. Standardize the soil EC value, and the expression is: ; and are the data mean and standard deviation respectively. Output the standardized data set ; The probability density function is: ; In the formula, is the weight of the k-th Gaussian distribution, satisfying ; is the probability density function of the k-th Gaussian distribution, and its form is: ; is the mean of the k-th Gaussian distribution, representing the center of a certain state, is the variance of the k-th Gaussian distribution, representing the distribution range of the state; Randomly initialize the initial value of , and calculate the probability that each data point belongs to the k-th Gaussian distribution: ; In the formula, is the responsibility weight of the i-th data point belonging to the k-th distribution. Update the GMM parameters according to , and the expression is: ; Calculate the total probability of the data point ​ , the expression is: ; The soil EC value stress anomaly index is defined as the probability that the point belongs to the non-normal distribution, and the expression is: ; YGF is the soil EC value stress anomaly index, is the probability that the point belongs to the normal distribution, and the expression is: ; In the formula, is the weight, mean, and standard deviation of the Gaussian distribution under normal conditions set according to historical data.

[0009] Preferably, in the pest and disease discrimination module, combining the characteristics of the fluctuation rate of pest and disease spread and the characteristics of soil EC value stress, the discrimination accuracy of similar symptoms of pests and diseases is evaluated. Specifically: Convert the pest and disease spread rate fluctuation index and the soil EC value stress anomaly index into a comprehensive feature vector, use the comprehensive feature vector as the input of the machine learning model, and use the machine learning model to predict the discrimination accuracy value label of similar symptoms of pests and diseases for each group of comprehensive feature vectors as the prediction target, and use minimizing the sum of the prediction errors of the discrimination accuracy value labels of all similar symptoms of pests and diseases as the training target to train the machine learning model until the sum of the prediction errors reaches convergence and stop the model training, and determine the discrimination accuracy value of similar symptoms of pests and diseases according to the model output result. Among them, the machine learning model is a polynomial regression model.

[0010] Preferably, in the diagnosis result classification module, the accuracy of the pest and disease diagnosis result is classified, and the pest and disease diagnosis result is classified into accurate diagnosis result, incomplete accurate diagnosis result, and inaccurate diagnosis result. Specifically: Compare the obtained discrimination accuracy value of similar symptoms of pests and diseases with the gradient standard threshold. The gradient standard threshold includes the first standard threshold and the second standard threshold, and the first standard threshold is less than the second standard threshold, and compare the discrimination accuracy value of similar symptoms of pests and diseases with the first standard threshold and the second standard threshold respectively; If the discrimination accuracy value of similar symptoms of pests and diseases is greater than the second standard threshold, it indicates that the discrimination accuracy of similar symptoms of pests and diseases is high. At this time, generate a high-accuracy discrimination signal and classify the diagnosis result as an accurate diagnosis result; If the discrimination accuracy value of similar symptoms of pests and diseases is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the discrimination accuracy of similar symptoms of pests and diseases is medium. At this time, generate a medium-accuracy discrimination signal and classify the diagnosis result as an incomplete accurate diagnosis result; If the discrimination accuracy value of similar symptoms of pests and diseases is less than the first standard threshold, it indicates that the discrimination accuracy of similar symptoms of pests and diseases is low. At this time, generate a low-accuracy discrimination signal and classify the diagnosis result as an inaccurate diagnosis result.

[0011] Preferably, in the optimization management module, for the diagnosis results of incomplete accuracy, the accuracy of distinguishing similar symptoms of pests and diseases within a fixed time period is further analyzed, and the knowledge graph and classification model are dynamically optimized according to the analysis results. Specifically: Within the fixed time period T, collect the key data of the diagnosis results of incomplete accuracy to form a data set D: ; where is the accuracy value of distinguishing similar symptoms of the i-th diagnosed pest and disease, is the fluctuation index of the expansion rate of the i-th diagnosed pest and disease, is the stress anomaly index of the soil EC value of the i-th diagnosis, is the actual disease label, and a comprehensive evaluation index for the accuracy of distinction within a fixed time period is defined , and the expression is: ; is the accuracy function of the diagnosis. According to and the actual label judge whether the diagnosis is correct: ; N is the number of samples with incomplete accuracy diagnosis within the time period T.

[0012] Preferably, define the weight of the node in the knowledge graph and the disease node , and update the weight based on the diagnosis data within the fixed time period T. The expression is: ; where , is the contribution prediction value of the node to the disease node in the i-th sample, and η is the learning rate, which is used to control the amplitude of weight update; Based on the comprehensive feature vector (WDC, YGF) and the actual label , retrain the classification model to minimize the prediction error of the accuracy of distinction: ; where is the objective function, is the accuracy value of distinction predicted by the classification model, and the model parameters θ are updated using the gradient descent method: ; is the learning rate, is the gradient of the objective function with respect to the model parameters θ, and the overall diagnosis accuracy index after dynamic optimization is defined , and the expression is: , the total number of samples in the N data set.

[0013] In the above technical solution, the technical effects and advantages provided by the present invention: 1. The present invention constructs a crop pest and disease diagnosis system based on a knowledge graph, overcoming the deficiencies of the prior art in distinguishing diseases with similar symptoms. By introducing the pest and disease expansion rate fluctuation index and the soil EC value stress anomaly index, and combining natural language processing and computer vision technologies, the system realizes the integration and dynamic analysis of multi-dimensional data. By optimizing the prediction of the discrimination accuracy value through a machine learning model, the diagnosis accuracy of diseases with similar symptoms is further improved. In addition, the system dynamically adjusts the node weights in the knowledge graph and continuously iterates and updates through the optimization of the classification model to ensure efficient and accurate disease discrimination capabilities in complex agricultural scenarios, thereby reducing the risk of misdiagnosis and providing scientific and reliable diagnostic basis for farmers.

[0014] 2. Through modular design, the system of the present invention can grade the accuracy of the diagnosis results and provide further optimization measures for incomplete accuracy and inaccurate results, making the diagnosis results more reliable. The dynamic optimization management module continuously updates the knowledge graph and the classification model using the diagnosis data within a fixed time period, significantly enhancing the system's adaptability to atypical symptoms such as newly emerging diseases and environmental stresses. Such a comprehensive and efficient pest and disease diagnosis system not only helps farmers accurately identify diseases, but also optimizes prevention and control measures, avoids the use of wrong pesticides, reduces crop losses and environmental pollution, and at the same time provides strong technical support for agricultural intelligent management. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0016] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiments, please refer to Figure 1As shown in the figure, the crop pest and disease diagnosis system based on a knowledge graph in this embodiment includes a data acquisition module, a data processing module, a knowledge graph construction module, a pest and disease differentiation module, a diagnosis result classification module, and an optimization management module; The data acquisition module is used to obtain the pest and disease description information, pest and disease image data, and environmental data of the crops, where the environmental data includes meteorological data and soil data; The data processing module is used to perform natural language processing on the pest and disease description information to extract symptom keywords, process the pest and disease image data through computer vision technology and extract the pest and disease expansion rate fluctuation characteristics, and obtain the soil EC value stress characteristics after processing the environmental data; The knowledge graph construction module is used to establish a knowledge graph based on pest and disease knowledge, where the knowledge graph includes pest and disease symptom nodes, environmental condition nodes, dynamic feature nodes, and the association relationships between the nodes; The pest and disease differentiation module is used to match the extracted symptom keywords with the symptom information stored in the knowledge graph, differentiate the pests and diseases according to the matching results, and evaluate the differentiation accuracy of the similar symptoms of the pests and diseases in combination with the pest and disease expansion rate fluctuation characteristics and the soil EC value stress characteristics; The diagnosis result classification module is used to classify the accuracy of the pest and disease diagnosis results according to the evaluation results, classify the pest and disease diagnosis results into accurate diagnosis results, incomplete accurate diagnosis results, and inaccurate diagnosis results, and perform corresponding processing; For the incomplete accurate diagnosis results, the optimization management module further analyzes the differentiation accuracy of the similar symptoms of the pests and diseases within a fixed time period, and dynamically optimizes the knowledge graph and the classification model according to the analysis results to improve the accuracy of crop pest and disease diagnosis.

[0019] In the data acquisition module, the user inputs text descriptions, such as the crop symptom information uploaded by farmers or technicians through mobile devices or Web platforms. The acquisition content includes: symptom keywords and specific descriptions related to diseases, such as: "There are irregular grayish-white spots on the rice leaves." "There are black dots the size of a needle tip on the leaves, which gradually expand." "The leaves turn yellow and the plant grows poorly." The user can input the crop variety, growth stage (such as tillering stage, heading stage), and the specific time and location where the disease occurs. Through natural language processing technology, the key information in the description (such as "grayish-white spots", "yellowing of the leaves") is extracted and stored in a structured manner for subsequent analysis.

[0020] Users upload images of crop diseases, such as photos taken by smartphones or drones. The collected content includes: images of disease symptoms on leaves, stems or fruits, such as: the shape, color, and edge characteristics of spots. The distribution pattern of diseases (such as concentrated at the leaf tip or randomly distributed). The overall texture change of the leaves (such as withered, curled, deformed). Field environment images: Collect the overall view of the field to understand the growth situation of the crops and the distribution characteristics in the field. Collection methods: Static image collection: High-definition photos taken and uploaded by users. Dynamic image collection: Continuous videos or multi-angle images taken by drones or agricultural robots. Image quality verification: Check the clarity, brightness, etc. of the images through computer vision technology to filter out low-quality images.

[0021] Environmental conditions play an important role in the occurrence and development of pests and diseases. Collecting environmental data can significantly improve the accuracy of the diagnostic system. Environmental data is divided into the following two parts: Meteorological data collection includes: Online meteorological data interface: Obtain real-time meteorological data from national or regional meteorological service platforms. Field sensors: Meteorological station equipment installed in farmland, including temperature and humidity sensors, rainfall gauges, etc. Collection content: Temperature: A key inducement for the occurrence of pests and diseases (such as high temperature leading to an accelerated reproduction of rice planthoppers). Humidity: High humidity is conducive to the spread of fungal diseases (such as rice blast). Rainfall: Monitor the impact of rain on soil humidity and disease spread. Wind speed and direction: Affect the spread direction of insect pests (such as the rice leaf folder). Format and standardize the real-time collected meteorological data. Associate the meteorological data with the environmental inducement conditions recorded in the pest and disease knowledge graph as an important basis for reasoning.

[0022] Soil data collection: Field soil sensors: Directly collect field soil data through Internet of Things devices. Manual measurement and experimental analysis: Send artificial samples to the laboratory for analysis (such as soil component testing). Collection content: Soil EC value (electrical conductivity): Reflects the degree of saline-alkali stress and helps distinguish between disease and non-disease inducements. Soil humidity: Affects the health of plant roots and the occurrence of diseases (such as root rot). Soil pH value: Detect whether there is an overly acidic or alkaline environment, which may cause nutrient deficiencies or saline-alkali stress. Soil nutrient components: Such as nitrogen, phosphorus, and potassium contents, to judge whether similar disease symptoms are caused by lack of fertilizer. Standardize the dimension of soil data through a standardization algorithm (such as unifying humidity into a percentage form). Associate with environmental factors related to crop diseases to identify potential stress characteristics.

[0023] In this application, the data collection module forms the basic input for system diagnosis by obtaining pest and disease description information, disease image data, and meteorological and soil environment data. The disease description information provides semantic features in text description, the image data provides visual features, and the environmental data provides key external conditions for disease reasoning.

[0024] A data processing module is used to perform natural language processing on the pest and disease description information to extract symptom keywords, process the pest and disease image data through computer vision technology and extract the fluctuation characteristics of the pest and disease expansion rate, and obtain the soil EC value stress characteristics after processing the environmental data.

[0025] Adopt natural language processing (NLP) technology to parse the text description information input by the user, extract symptom keywords, and convert unstructured text into structured data that can be used for reasoning. The processing steps include: Text preprocessing: Remove irrelevant characters such as stop words and punctuation marks, and standardize the text format. Example input: "Irregular grayish-white spots appear on the rice leaves." Keyword extraction: Use lexical analysis and word segmentation technology to extract symptom-related keywords. Example output: {Crop type: rice, Disease location: leaves, Symptom characteristics: irregular grayish-white spots}. Symptom association matching: Use semantic matching technology to match the extracted keywords with the standardized symptom thesaurus in the knowledge graph to identify potential disease nodes. For example, the keyword "grayish-white spots" matches the symptom nodes of "rice blast" and "bacterial leaf blight". Output result: Structured symptom keyword data, providing basic input for disease reasoning in the knowledge graph.

[0026] Utilize computer vision technology to process the disease images uploaded by the user, extract the visual characteristics of pests and diseases, including dynamic information such as the fluctuation characteristics of the expansion rate. The processing steps include: Image preprocessing: Perform operations such as denoising, brightness adjustment, and image cropping on the input image to ensure that the image quality is suitable for analysis. Disease feature extraction: Use a convolutional neural network (CNN) model to extract the key visual features in the image: Spot color: grayish-white, brown or black. Spot shape: regularity, irregularity. Spot edge: blurred or clear. Example output: {Color: grayish-white, Shape: irregular, Edge: blurred}. Fluctuation characteristics of the disease expansion rate: When the user uploads disease images of the same plant multiple times, the system compares the images at different time periods and calculates the disease expansion rate. Use image segmentation technology (such as Mask R-CNN) to identify the diseased area and calculate the change in the diseased area: Expansion rate = (current diseased area - previous diseased area) / time interval. Example output: Expansion rate = 5 cm² / day. Output result: The extracted disease visual feature data and the fluctuation characteristics of the expansion rate provide a time dynamic basis for distinguishing highly similar symptoms.

[0027] Standardize and extract features from the collected environmental data (such as soil data), with a focus on identifying the stress characteristics of soil EC values, which serve as an important distinguishing indicator between diseases and non-diseases (such as saline-alkali stress). The processing steps include: Data standardization: Denoise and standardize the original soil data (such as the electrical conductivity EC value) collected by the sensor. Example: The original EC value = 2.5 mS / cm, which is converted into a comparable indicator after standardization. Stress feature extraction: Set the normal range threshold of the soil EC value (such as 0.5 mS / cm - 2.0 mS / cm). When the EC value exceeds the normal range, it is marked as a saline-alkali stress feature. Example output: Soil EC value = 2.8 mS / cm → Marked as "saline-alkali stress". Soil humidity = 30% → Assist in confirming whether symptoms are caused by water stress. Association of environmental data and visual data: Integrate the stress characteristics of soil EC values with the characteristics of disease images to determine whether the symptoms are caused by saline-alkali stress rather than disease infection. Output result: Standardized soil EC value data and stress characteristics provide an environmental basis for the system to rule out non-disease factors.

[0028] After analyzing the extracted fluctuation characteristics of the pest and disease expansion rate, a pest and disease expansion rate fluctuation index is generated. The method for obtaining the pest and disease expansion rate fluctuation index is as follows: Obtain the disease expansion rate data for different time periods and construct two time series: : The reference expansion rate sequence, representing the standard disease expansion rate pattern (such as the typical expansion trend recorded in the knowledge graph). : The currently collected disease expansion rate sequence (such as the rate fluctuation obtained from actual data). Each rate point represents the change rate of the diseased area in a certain time period. The lengths of the two sequences may be different, i.e., m≠n, and the time points may not be aligned. For any two points , calculate the local distance: ; In the formula, is the point and 's absolute distance.

[0029] Construct a cost matrix D, which is an m×n matrix. Each element D(i,j) represents the minimum cumulative distance of the path from the starting point (1,1) to the point (i,j). The initial cost is: , ; , ; Fill the matrix sequentially through dynamic programming and cumulatively calculate the minimum matching cost between points.

[0030] Define the optimal path as a sequence of point pairs, and the condition is: represents the starting point; Indicates the end point. The path is continuous and cannot jump. The total cost of the path is: ; Calculate the fluctuation index of the pest and disease spread rate. The expression is: ; In the formula, is the path length (the number of matching points), and WDC is the fluctuation index of the pest and disease spread rate.

[0031] When the fluctuation index of the pest and disease spread rate is larger, it indicates that the difference between the current disease spread rate and the standard spread rate pattern is more significant, that is, the dynamic feature matching degree of the time series is lower. Such a large fluctuation index may indicate that: The current disease may have a significant difference in the spread trend from the reference standard diseases (such as rice blast and bacterial leaf blight), and the system has a high accuracy in distinguishing highly similar symptoms.

[0032] However, if the system knowledge graph coverage is insufficient, this large difference may also be due to abnormal spread caused by newly emerging diseases or non-disease factors (such as environmental stress). Therefore, in the case of a large fluctuation index, other data (such as soil EC value, image features) need to be combined for comprehensive analysis to ensure the reliability of the diagnosis result.

[0033] When the fluctuation index of the pest and disease spread rate is smaller, it indicates that the dynamic features of the current disease spread rate are highly similar to the standard spread pattern, and the time series matching degree is higher. This usually means that: The current disease is very likely to belong to the same category as the standard disease, and the system may have a lower accuracy in distinguishing similar symptoms because the similarity of the spread rate may mask the differences between diseases.

[0034] In this case, it is necessary to further rely on other features (such as the edge sharpness and color features of the symptom image) to improve the discrimination ability, otherwise it is easy to lead to misdiagnosis. For example, rice blast and bacterial leaf blight may show consistency in the spread rate, but it may be easier to distinguish through the lesion morphology or soil characteristics.

[0035] After analyzing the extracted soil EC value stress characteristics, a soil EC value stress anomaly index is generated. The acquisition method of the soil EC value stress anomaly index is: Collect the sampling data set of the soil EC value: input data , each represents the soil EC value of a sampling point, and the data may contain different states (such as normal, mild stress, severe stress).

[0036] Normalize or standardize the soil EC value. The expression is: ; and are the data mean and standard deviation respectively, and the standardized data set is output ; The GMM assumes that the data follows a mixture model of K Gaussian distributions, and the probability density function is: ; In the formula, is the weight of the k-th Gaussian distribution, satisfying ; is the probability density function of the k-th Gaussian distribution, and its form is: ; is the mean of the k-th Gaussian distribution, representing the center of a certain state, is the variance of the k-th Gaussian distribution, representing the distribution range of the state.

[0037] Randomly initialize the initial value of, and calculate the probability that each data point belongs to the k-th Gaussian distribution: ; In the formula, is the responsibility weight of the i-th data point belonging to the k-th distribution. According to update the GMM parameters, and the expression is: ; Calculate the total probability of the data point , and the expression is: ; The soil EC value stress anomaly index is defined as the probability that the point belongs to an abnormal distribution, and the expression is: ; YGF is the soil EC value stress anomaly index, is the probability that the point belongs to a normal distribution, and the expression is: ; In the formula, are the weight, mean, and standard deviation of the Gaussian distribution under normal conditions set according to historical data.

[0038] When the soil EC value stress anomaly index is larger, it indicates that the current soil salinity stress level significantly deviates from the normal range. This shows that plants may exhibit symptoms similar to those of pests and diseases (such as leaf spots, yellowing, etc.) due to non-disease reasons (such as salinity stress or environmental stress). In this case, the accuracy of distinguishing symptoms similar to those of pests and diseases is higher because the system can identify non-disease causes through a high anomaly index and exclude incorrect disease diagnoses, avoiding misusing pest control measures such as pesticides. Therefore, a larger anomaly index helps improve the ability of the diagnostic system to distinguish between disease and non-disease causes.

[0039] When the stress anomaly index of soil EC value is smaller, it indicates that the soil salinity level is close to the normal range, and the plant symptoms are likely caused by pests and diseases rather than soil stress factors. At this time, the system needs to rely on other features (such as the expansion rate, lesion morphology) to distinguish the disease types with similar symptoms. Due to the exclusion of soil stress, the accuracy of distinguishing pest and disease symptoms may decrease, especially between highly similar diseases (such as rice blast and bacterial leaf blight). Therefore, when the anomaly index is small, more disease feature analyses need to be combined to avoid misdiagnosis or ambiguous diagnosis.

[0040] The knowledge graph construction module is used to establish a knowledge graph based on pest and disease knowledge. The knowledge graph includes pest and disease symptom nodes, environmental condition nodes, dynamic feature nodes, and the association relationships between each node.

[0041] The nodes in the knowledge graph are the basic units of pest and disease related knowledge. Different node types represent multi-dimensional features of the occurrence of pests and diseases and their related factors: Definition of pest and disease symptom nodes: Describe the typical plant symptoms caused by pests and diseases. Data sources: Agricultural literature, expert knowledge bases, historical disease case data. Main attributes: Symptom description (text): Such as "grayish-white spots on leaves", "wilted leaf margins". Typical features (numerical): Such as lesion area, color, edge features. Disease types: Such as rice blast, bacterial leaf blight, rice planthopper damage, etc. Affected parts: Such as leaves, stems, fruits, etc. Example node: Symptom node: "Grayish-white spots on leaves" Disease type: "Rice blast" Part: "Leaves" Features: "Irregular spots, fast expansion rate" Definition of environmental condition nodes: Describe the external environmental factors related to the occurrence of pests and diseases, including meteorological conditions and soil conditions. Data sources: Meteorological data interfaces, sensor monitoring data, experimental data. Main attributes: Meteorological conditions: Temperature, humidity, rainfall, wind speed, etc. Soil conditions: EC value, pH value, humidity, nutrient content, etc. Inducing conditions: The triggering effect of the environment on diseases, such as high humidity promoting the reproduction of fungi. Example node: Environmental condition node: "High humidity (>80%)" Induced disease: "Rice blast" Environmental condition node: "Excessively high soil EC value (>2.5 mS / cm)" Induced symptom: "Yellowing of leaves due to saline-alkali stress" Definition of dynamic feature nodes: Describe the dynamic features of pest and disease symptoms changing over time. Data sources: Time series data, observations of disease expansion experiments. Main attributes: Expansion rate: Such as the growth rate of lesion area (unit: cm² / day). Change pattern: Such as the disease expansion showing linear growth, explosive growth, or a stable stage. Time nodes: Such as "peak expansion period", "symptom remission period", etc. Example node: Dynamic feature node: "Expansion rate 5 cm² / day" Disease: "Rice blast" Stage: "Mid-term expansion stage" Knowledge graphs establish associations through semantic relationships between nodes, reflecting the internal connections and causal relationships of pest and disease-related factors: Relationship between symptoms and diseases: The symptoms caused by a certain pest or disease and their manifestations. Example: Relationship type: "causes symptoms" Rice blast → causes symptoms → Grayish-white spots; Rice planthopper damage → causes symptoms → Leaf yellowing. Relationship between diseases and the environment: The occurrence or spread of a certain pest or disease is affected by specific environmental conditions. Example: Relationship type: "affected by" Rice blast → affected by → High humidity; Salinity stress → affected by → Excessive soil EC value. Relationship between dynamic characteristics and symptoms: The dynamic manifestation of a disease symptom over time. Example: Relationship type: "dynamically manifested as" Grayish-white spots → dynamically manifested as → Expansion rate of 5 cm² / day.

[0042] Through natural language processing (NLP) techniques, extract semantic information on diseases, symptoms, environment, and dynamic characteristics from agricultural literature and reports. Extract feature values through image processing and environmental sensor data analysis. Use a graph database (such as Neo4j) or semantic technologies (such as RDF / OWL) to establish the graph structure. Define node types, attributes, and association relationships. Perform consistency processing on multi-source data (literature, sensors, image analysis results) to eliminate redundant or conflicting nodes. Add new nodes and relationships based on new disease data or user feedback.

[0043] The pest and disease differentiation module is used to match the extracted symptom keywords with the symptom information stored in the knowledge graph, differentiate pests and diseases according to the matching results, and evaluate the differentiation accuracy of similar symptoms of pests and diseases in combination with the fluctuation characteristics of the pest and disease expansion rate and the stress characteristics of the soil EC value.

[0044] Match the symptom keywords with the symptom information in the knowledge graph, which are the symptom keywords extracted from the description information and image data uploaded by the user. Semantically match the symptom keywords with the symptom nodes and their characteristic attributes in the knowledge graph. The matching rules can be based on keyword similarity, semantic similarity (such as Word2Vec or BERT embeddings), graph path weights, etc. Example: Input keywords: "Grayish-white spots", "Leaf yellowing". Matched knowledge graph nodes: Grayish-white spots → Rice blast. Leaf yellowing → Rice planthopper damage. Output a list of candidate diseases or pests according to the matching degree, and calculate the matching weights of each disease. Combine dynamic characteristics (such as expansion rate) and environmental conditions to re-rank the confidence levels of the candidate diseases and determine the final result.

[0045] Based on natural language processing (NLP) technology, structured keywords are extracted from the description entered by the user. For example, the input description is: "Irregular gray-white spots appear on rice leaves and gradually expand." Extract keywords: {"gray-white spots", "irregular", "gradually expand"}. Using computer vision technology, features (such as lesion color, shape, and distribution) are extracted from the disease image. For example: color = gray-white, edge = fuzzy, expansion rate = 5 cm² / day. In the symptom nodes of the knowledge graph, nodes that fully or partially match the extracted keywords are retrieved. Example: Symptom keyword: "gray-white spots" knowledge graph matching nodes: {"rice blast", "bacterial leaf spot"}. Use deep learning embedding models (such as BERT, Word2Vec) to calculate the semantic similarity between the input keywords and the symptom description in the knowledge graph: Q is the extracted symptom keyword, and S is the symptom description stored in the knowledge graph. Output a list of possible pest and disease candidates by sorting by similarity.

[0046] The matching process combines dynamic features and environmental conditions: Dynamic feature matching: Compare the dynamic curves of expansion rate or morphological change. Environmental condition matching: Check whether the disease-inducing conditions (such as humidity > 80%, high temperature, etc.) are met.

[0047] Taking into account the symptom similarity, dynamic feature similarity, and environmental matching degree, the matching weight of each pest and disease is calculated: ; Where α, β, and γ are weight factors that adjust the importance of symptoms, dynamic characteristics, and environmental conditions. is the comprehensive matching weight of candidate disease i. Sort the candidate diseases and select the disease with the highest weight as the diagnosis result. Output confidence , the expression is: Confidence is used to explain the reliability of the diagnosis results. If the highest confidence is lower than the set threshold, a list of candidate diseases and their confidence levels is output, prompting the user for further confirmation.

[0048] Candidate diseases are re-ranked based on dynamic features such as expansion rate and morphological changes to improve classification accuracy. For example, if the expansion rate is 5 cm² / day, "rice blast" is prioritized (its dynamic features are highly consistent). Environmental condition data (such as humidity and soil EC values) are used to exclude diseases that do not meet the inducing conditions. For example, if the soil EC value is >2.5 mS / cm, salinity stress is more likely to be diagnosed than disease infection. Based on user feedback on diagnostic results (such as confirmation of disease type), the weights of knowledge graph nodes are dynamically updated to continuously optimize the matching algorithm.

[0049] The application scenarios include: single disease diagnosis, such as directly diagnosing rice blast or rice planthopper damage; differentiating similar symptoms, such as differentiating rice blast and bacterial leaf blight, and combining the expansion rate and environmental conditions to give more accurate diagnostic results; dynamic change monitoring, differentiating disease and non-disease stresses by combining dynamic characteristics in continuous monitoring.

[0050] Convert the pest and disease expansion rate fluctuation index and the soil EC value stress anomaly index into a comprehensive feature vector. Use the comprehensive feature vector as the input of a machine learning model. The machine learning model takes the prediction of the differentiation accuracy value label of similar pest and disease symptoms for each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of the prediction errors of the differentiation accuracy value labels of all similar pest and disease symptoms as the training target. Train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the differentiation accuracy value of similar pest and disease symptoms according to the model output result, where the machine learning model is a polynomial regression model.

[0051] The method for obtaining the differentiation accuracy value of similar pest and disease symptoms is: obtain the corresponding function expression from the comprehensive feature vector training data of the trained machine learning model: ; in the formula, is the output function of the model, WDC is the pest and disease expansion rate fluctuation index, YGF is the soil EC value stress anomaly index, and ASD is the differentiation accuracy value of similar pest and disease symptoms.

[0052] The diagnosis result classification module classifies the accuracy of the pest and disease diagnosis result according to the evaluation result, divides the pest and disease diagnosis result into accurate diagnosis result, incomplete accurate diagnosis result and inaccurate diagnosis result, and performs corresponding processing.

[0053] Compare the obtained differentiation accuracy value of similar pest and disease symptoms with the gradient standard thresholds. The gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the differentiation accuracy value of similar pest and disease symptoms with the first standard threshold and the second standard threshold respectively; If the differentiation accuracy value of similar pest and disease symptoms is greater than the second standard threshold, it indicates that the differentiation accuracy of similar pest and disease symptoms is high. At this time, generate a high-accuracy differentiation signal and classify the diagnosis result as an accurate diagnosis result; If the differentiation accuracy value of similar pest and disease symptoms is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the differentiation accuracy of similar pest and disease symptoms is medium. At this time, generate a medium-accuracy differentiation signal and classify the diagnosis result as an incomplete accurate diagnosis result; If the accuracy value of differentiating similar symptoms of pests and diseases is less than the first standard threshold, it indicates that the accuracy of differentiating similar symptoms of pests and diseases is low. At this time, a low-accuracy differentiation signal is generated, and the diagnostic result is classified as an inaccurate diagnostic result.

[0054] Processing of accurate diagnostic results: When the accuracy value of differentiating similar symptoms of pests and diseases ASD > the second standard threshold, it indicates that the diagnostic result has a high credibility, and the system has a strong ability to differentiate highly similar symptoms. Directly output the diagnostic result: directly output the diagnosed disease name, cause of disease, disease characteristics, and prevention suggestions to the user. Example: Diagnostic result: "Rice blast". Prevention suggestion: It is recommended to use a fungicide (tricyclazole), and the spraying concentration is 200 - 300 times liquid, or strengthen field ventilation to reduce humidity. Provide a detailed reasoning path: Display the reasoning process of the diagnosis through the knowledge graph, including: the matching degree between the symptom keywords and the symptom nodes of the knowledge graph. The similarity analysis of dynamic features (such as the expansion rate) and standard features. The influence weight of environmental conditions (such as soil EC value) on the diagnosis. Let the user understand the diagnostic basis and increase the trust. Result storage and optimization: Store the diagnostic result as sample data for further optimizing the model. Record successful cases to enhance the weight assignment and correlation analysis ability of the knowledge graph.

[0055] Processing of inaccurate diagnostic results: ASD < the first standard threshold, indicating that the diagnostic result has a low credibility, possibly due to: atypical symptoms or insufficient data. The knowledge graph coverage of the system is insufficient and cannot handle newly emerging diseases or non-disease factors (such as stress). The processing solutions include: Mark as diagnostic failure or unknown disease: Clearly inform the user that the diagnostic result is inaccurate and may be affected by unknown diseases or complex stress. Example: Prompt message: "The system cannot accurately determine the type of disease. Please take the following preliminary measures." Provide conventional prevention and control measures: According to the symptom characteristics, recommend that the user take general disease prevention and environmental regulation measures: Strengthen field management: such as improving ventilation and reducing humidity. Improve the soil environment: such as reducing the soil EC value by irrigation flushing under saline-alkali stress. Use broad-spectrum prevention and control measures: such as selecting broad-spectrum fungicides or biological control programs. Store the symptoms and environmental data that have not been successfully matched as "abnormal samples" and trigger the following mechanisms: Knowledge graph expansion: Add new disease nodes or stress feature nodes to enhance the knowledge coverage. Model retraining: Update the machine learning model with abnormal sample data to optimize its diagnostic ability. If the diagnostic result is an unknown disease, upload the relevant data (such as images, environmental characteristics, dynamic characteristics) to the agricultural diagnosis platform for expert manual review.

[0056] Optimize the management module. For incompletely accurate diagnostic results, further analyze the accuracy of differentiating similar symptoms of pests and diseases within a fixed time period, and dynamically optimize the knowledge graph and classification model according to the analysis results to improve the accuracy of crop pest and disease diagnosis.

[0057] Collect the key data of the incomplete accuracy diagnosis results within a fixed time period T to form a data set D: ; where is the accuracy value for distinguishing similar symptoms of pests and diseases in the i-th diagnosis, is the fluctuation index of the spread rate of pests and diseases in the i-th diagnosis. is the stress anomaly index of the soil EC value in the i-th diagnosis, is the actual disease label, which is used to verify the accuracy of the diagnosis results (obtained through subsequent follow-up consultations or expert feedback).

[0058] Define the comprehensive evaluation index of the distinguishing accuracy within a fixed time period , and the expression is: ; is the accuracy function of the diagnosis. According to and the actual label judge whether the diagnosis is correct: ; N is the number of incomplete accuracy diagnosis samples within the time period T.

[0059] Analyze the key features (deficiencies in symptom nodes, environmental condition nodes, and dynamic feature nodes) that lead to misdiagnosis in the incomplete accuracy diagnosis samples, and dynamically adjust the structure and weights of the knowledge graph: According to expert feedback or system follow-up consultation results, supplement the missing disease, symptom, or environmental condition nodes in the knowledge graph. Reassign the weights of the node relationships with a higher misdiagnosis frequency to improve the matching accuracy.

[0060] Define the weight of the node in the knowledge graph and the disease node , and update the weight based on the diagnosis data within the fixed time period T. The expression is: ; where , is the contribution prediction value of the node to the disease node in the i-th sample, and η is the learning rate, which is used to control the weight update amplitude.

[0061] Based on the comprehensive feature vector (WDC, YGF) and the actual label , retrain the classification model to minimize the prediction error of the distinguishing accuracy: ; where is the objective function (loss function), is the distinguishing accuracy value predicted by the classification model. Use the gradient descent method to update the model parameters θ: ; is the learning rate, is the gradient of the objective function with respect to the model parameter θ. Define the overall diagnostic accuracy index after dynamic optimization , and the expression is: ; where The closer it is to 1, the higher the diagnostic accuracy of the system for pests and diseases with similar symptoms.

[0062] In this embodiment, the precise diagnosis and dynamic optimization of crop pests and diseases are achieved through the collaborative work of multiple modules. The data acquisition module obtains pest and disease description information, image data, and environmental data such as meteorology and soil; the data processing module extracts symptom keywords, the fluctuation characteristics of the pest and disease expansion rate, and the soil EC value stress characteristics through natural language processing and computer vision technologies; the knowledge graph construction module builds a knowledge graph including symptom nodes, environmental condition nodes, dynamic feature nodes, and their association relationships based on pest and disease knowledge; the pest and disease differentiation module matches the extracted symptom keywords with the knowledge graph and evaluates the differentiation accuracy of similar symptoms of pests and diseases in combination with the expansion rate and soil stress characteristics; the diagnostic result classification module divides the diagnostic results into accurate diagnostic results, incomplete accurate diagnostic results, and inaccurate diagnostic results according to the differentiation accuracy, and takes targeted measures; the optimization management module further analyzes the incomplete accurate diagnostic results, dynamically optimizes the knowledge graph and the classification model to continuously improve the accuracy of the diagnostic system.

[0063] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0064] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains a set of one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0065] As described above, the specific implementation manners of the present application are only described, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application.

Claims

1. A crop pest and disease diagnosis system based on a knowledge graph, characterized in that: It includes a data acquisition module, a data processing module, a knowledge graph construction module, a pest and disease differentiation module, a diagnosis result classification module, and an optimization management module; The data acquisition module is used to obtain the pest and disease description information, pest and disease image data, and environmental data of crops, and the environmental data includes meteorological data and soil data; The data processing module is used to perform natural language processing on the pest and disease description information to extract symptom keywords, process the pest and disease image data through computer vision technology and extract the fluctuation characteristics of the pest and disease expansion rate, and obtain the soil EC value stress characteristics after processing the environmental data; The knowledge graph construction module is used to establish a knowledge graph based on pest and disease knowledge, and the knowledge graph includes pest and disease symptom nodes, environmental condition nodes, dynamic feature nodes, and the association relationships between the nodes; The pest and disease differentiation module is used to match the extracted symptom keywords with the symptom information stored in the knowledge graph, differentiate the pests and diseases according to the matching results, and evaluate the differentiation accuracy of similar pest and disease symptoms in combination with the fluctuation characteristics of the pest and disease expansion rate and the soil EC value stress characteristics; The diagnosis result classification module classifies the accuracy of the pest and disease diagnosis results according to the evaluation results, divides the pest and disease diagnosis results into accurate diagnosis results, incomplete accurate diagnosis results, and inaccurate diagnosis results, and performs corresponding processing; For the incomplete accurate diagnosis results, the optimization management module further analyzes the differentiation accuracy of similar pest and disease symptoms within a fixed time period, and dynamically optimizes the knowledge graph and the classification model according to the analysis results to improve the accuracy of crop pest and disease diagnosis.

2. The crop pest and disease diagnosis system based on a knowledge graph according to claim 1, wherein: In the data processing module, after analyzing the extracted fluctuation characteristics of the pest and disease expansion rate, a pest and disease expansion rate fluctuation index is generated. The method for obtaining the pest and disease expansion rate fluctuation index is as follows: Obtain the disease expansion rate data for different time periods and construct two time series: is the reference expansion rate sequence, representing the standard disease expansion rate pattern; is the disease expansion rate sequence currently collected. Each rate point represents the change rate of the lesion area in a certain time period. For any two points , calculate the local distance: ; where is the absolute distance between point and ; Construct the cost matrix D. D is an m×n matrix, where each element D(i,j) represents the minimum cumulative distance of the path from the starting point (1,1) to the point (i,j). The initial cost is as follows: , ; , ; Fill the matrix sequentially through dynamic programming, and cumulatively calculate the minimum matching cost between each point; Define the optimal path is a sequence of point pairs that satisfy the conditions: represents the starting point; represents the ending point. The path is continuous and cannot jump. The total cost of the path is: ; Calculate the pest and disease spread rate fluctuation index, and the expression is: ; In the formula, is the path length, and WDC is the pest and disease spread rate fluctuation index.

3. The crop pest and disease diagnosis system based on a knowledge graph according to claim 2, characterized in that: In the data processing module, after analyzing the extracted soil EC value stress characteristics, a soil EC value stress anomaly index is generated. The method for obtaining the soil EC value stress anomaly index is as follows: Sampling dataset for collecting soil EC values: input data , each represents the soil EC value of a sampling point. The soil EC value is standardized, and the expression is: ; and are the data mean and standard deviation respectively, and the standardized dataset is output; The probability density function is: ; In the formula, is the weight of the k-th Gaussian distribution, satisfying ; is the probability density function of the k-th Gaussian distribution, and its form is: ; is the mean of the k-th Gaussian distribution, representing the center of a certain state, is the variance of the k-th Gaussian distribution, representing the distribution range of the state; Random initialization Initialize the initial value, and calculate the probability that each data point belongs to the k-th Gaussian distribution: ; In the formula, is the responsibility weight that the i-th data point belongs to the k-th distribution. According to Update the GMM parameters, and the expression is: ; Calculate the total probability of the data point as , and the expression is: ; The soil EC value stress anomaly index is defined as the probability that the point belongs to the abnormal distribution, and the expression is: ; YGF is the soil EC value stress anomaly index, is the probability that the point belongs to the normal distribution, and the expression is: ; In the formula, is the weight, mean, and standard deviation of the Gaussian distribution under the normal state set according to historical data.

4. The crop pest and disease diagnosis system based on a knowledge graph according to claim 3, characterized in that: In the pest and disease differentiation module, the differentiation accuracy of similar pest and disease symptoms is evaluated in combination with the fluctuation characteristics of the pest and disease expansion rate and the soil EC value stress characteristics. Specifically: The pest and disease expansion rate fluctuation index and the soil EC value stress anomaly index are converted into a comprehensive feature vector, and the comprehensive feature vector is used as the input of a machine learning model. The machine learning model takes predicting the differentiation accuracy value label of similar pest and disease symptoms for each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of the prediction errors of the differentiation accuracy value labels of all similar pest and disease symptoms as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence and then the model training is stopped. The differentiation accuracy value of similar pest and disease symptoms is determined according to the model output result. Among them, the machine learning model is a polynomial regression model.

5. The crop pest and disease diagnosis system based on a knowledge graph according to claim 4, characterized in that: In the diagnosis result classification module, the accuracy of the pest and disease diagnosis results is classified, and the pest and disease diagnosis results are divided into accurate diagnosis results, incomplete accurate diagnosis results, and inaccurate diagnosis results. Specifically: Compare the obtained discrimination accuracy value of the similar symptoms of pests and diseases with the gradient standard thresholds, where the gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and compare the discrimination accuracy value of the similar symptoms of pests and diseases with the first standard threshold and the second standard threshold respectively; If the discrimination accuracy value of the similar symptoms of pests and diseases is greater than the second standard threshold, it indicates that the discrimination accuracy of the similar symptoms of pests and diseases is high. At this time, generate a high-accuracy discrimination signal and classify the diagnosis result as an accurate diagnosis result; If the discrimination accuracy value of the similar symptoms of pests and diseases is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the discrimination accuracy of the similar symptoms of pests and diseases is medium. At this time, generate a medium-accuracy discrimination signal and classify the diagnosis result as an incomplete-accuracy diagnosis result; If the discrimination accuracy value of the similar symptoms of pests and diseases is less than the first standard threshold, it indicates that the discrimination accuracy of the similar symptoms of pests and diseases is low. At this time, generate a low-accuracy discrimination signal and classify the diagnosis result as an inaccurate diagnosis result.

6. The crop pest and disease diagnosis system based on a knowledge graph according to claim 1, characterized in that: In the optimization management module, for the incomplete-accuracy diagnosis result, further analyze the discrimination accuracy of the similar symptoms of pests and diseases within a fixed time period, and dynamically optimize the knowledge graph and classification model according to the analysis results. Specifically: Collect the key data of the incomplete accuracy diagnosis results within a fixed time period T to form a data set D: ; where is the accuracy value for distinguishing similar symptoms of the pests and diseases diagnosed in the i-th case, is the fluctuation index of the spread rate of the pests and diseases diagnosed in the i-th case, is the stress anomaly index of the soil EC value for the i-th diagnosis, is the actual disease label, and define a comprehensive evaluation index for the accuracy of discrimination within a fixed time period , and the expression is: ; is the accuracy function of the diagnosis. According to and the actual label judge whether the diagnosis is correct: ; N is the number of incomplete accuracy diagnosis samples within the time period T.

7. The crop pest and disease diagnosis system based on a knowledge graph according to claim 6, characterized in that: Define nodes in the knowledge graph The weight of the disease node is updated based on the diagnostic data within a fixed time period T, and the expression is: ; In the formula, ; where , is the predicted contribution value of the node in the i-th sample to the disease node , and η is the learning rate used to control the magnitude of weight update; Based on the comprehensive feature vectors (WDC, YGF) and the actual labels , retrain the classification model to minimize the prediction error of discrimination accuracy: ; where is the objective function, is the discrimination accuracy value predicted by the classification model, and the model parameters θ are updated using the gradient descent method: ; is the learning rate, is the gradient of the objective function with respect to the model parameters θ, and the overall diagnostic accuracy index after dynamic optimization is defined as , and the expression is: , where N is the total number of samples in the data set.

Images