Forestry pest and disease prediction method and system based on big data

Abstract

The invention discloses a forestry pest and disease prediction method and system based on big data, and the method comprises the steps: obtaining a target monitoring data set of each block of a forest region, and enabling the target monitoring data set to comprise a target image, an environment parameter and forest region information; performing pest and disease damage area detection on the target image to obtain a forest region pest and disease damage feature set; predicting the forest region pest and disease feature set based on a forest region diffusion detection model to obtain a diffusion state prediction value; constructing a forestry disease and pest knowledge map, detecting the morbidity probability of different types of trees, and optimizing and generating forest disease and pest degree grades; potential causes of the forest area are analyzed according to the degree grade of forest diseases and insect pests, the accuracy and efficiency of disease edge detection are remarkably improved, and real-time recognition and early warning of the diseases and insect pests are achieved.

Description

Technical Field

[0001] This invention relates to the field of disease and pest diagnosis technology, and in particular to a method and system for predicting forestry diseases and pests based on big data. Background Technology

[0002] The prevention and control of forest pests and diseases is a key task in the development of forestry. Traditional identification of forest pests and diseases mainly relies on experts to conduct on-site inspections, which is inefficient and prone to misjudgment. The big data-based forest pest and disease identification and prediction method has the advantages of high accuracy and high efficiency, providing an effective technical means for the scientific prevention and control of forest pests and diseases.

[0003] A Chinese invention with application number 202510759296.3 discloses a method and system for predicting forest pests and diseases based on artificial intelligence. The main method of this invention includes: acquiring a dynamic monitoring data set of the target forest area, extracting features from the set to obtain a set of forestry environmental features for each monitoring node, establishing a pre-trained pest and disease prediction model, performing abnormal parameter analysis on the set of forestry environmental features to generate a set of pest and disease prediction parameters for the monitoring node, which represents the probability of pest and disease occurrence and the level of impact range, performing dynamic prediction matching operation based on the pest and disease prediction parameters to form a heat map of pest and disease distribution in the target forest area, generating a pest and disease priority strategy, and finally blending the heat map and priority strategy to generate a set of optimized instructions for forest area prevention and control.

[0004] The above methods do not integrate multi-source analysis of environment, pathogens and tree species, fail to construct a knowledge graph of forestry pests and diseases, cannot distinguish between correlations and actual causes, and the models only make independent predictions for a single detection node. They fail to consider spatial propagation patterns and time cumulative effects, and lack spatiotemporal correlation. Summary of the Invention

[0005] The technical problem solved by this invention is that: without the joint analysis of environment, pathogens and tree species, it fails to distinguish the correlation between pests and tree species and the true cause of disease. Existing prediction models mostly use environmental parameters as static or discrete input features, which makes it difficult to reflect the continuous influence of the environment on the spread rate and direction of pests and diseases.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Firstly, a big data-based method for predicting forestry pests and diseases includes: Step S1: Obtain target monitoring data sets for each block in the forest area, wherein the target monitoring data sets include target images, environmental parameters, and forest area information; Step S2: Detect pest and disease areas in the target image to obtain a set of forest pest and disease features; Step S3: Based on the forest area diffusion detection model, predict the feature set of forest area pests and diseases to obtain the predicted value of diffusion status; Step S4: Construct a knowledge graph of forestry diseases and pests, detect the probability of disease occurrence in different types of trees, and optimize the generation of forest disease and pest severity levels; Step S5: Analyze the potential causes of diseases in the forest area based on the severity level of forest pests and diseases.

[0007] Preferably, step S1 specifically includes: Step S11: Divide the various blocks in the forest area into several blocks according to tree species, and use drone aerial photography technology to take several target images of several blocks. The target images are used to capture the current state of the trees and the state of pests and diseases. Step S12: Install ground sensors in several blocks to obtain environmental parameters of the forest area. The ground sensors include temperature and humidity sensors and soil monitoring sensors. The environmental parameters include temperature, humidity, soil moisture content and soil pH. Step S13: Obtain forest area information based on historical forest area records. The forest area information includes tree species, tree height, trunk diameter, tree age, and historical disease records. Step S14: Save the target image, environmental parameters, and forest area information into a target monitoring data set according to n blocks.

[0008] Preferably, the specific method for detecting pest and disease areas in the target image includes: The target image is noise-reducing and binary grayscale conversion is performed using median filtering, and the resolution is adjusted to W×H. The target image is divided into a grid. Starting from the top left corner of the target image, the pixel gradient magnitudes in the horizontal and vertical directions are calculated. The pixel gradient magnitudes within the grid are statistically analyzed, and the average gradient magnitude of each grid is calculated. The average gradient magnitudes are iterated, and the two adjacent grids with the largest differences are found as the starting and ending points of the pest and disease region. Pest and disease region detection is set up, which includes a first detection and a second detection. The coordinates of the starting and ending points are found on the target image, and a sampling interval is set. The first detection is used to obtain a first fitted contour map. The pixel grayscale values ​​of the target image are queried, starting from the starting point position, to set... The pixel grayscale values ​​of the target image are traversed at a fixed unit angle. When the pixel grayscale value matches the pixel fit of the diseased tree, the feature points of the diseased tree are extracted and the location of the feature points is marked until the coordinates of the last point are returned. The enclosed area is fitted into a first fitted contour map, and the marked points of the diseased tree area are generated. The proportion of the area of ​​the diseased tree area to the area of ​​the entire target image is calculated. The second detection is used to obtain the second fitted contour map. Pixels of the grid area on the target image are randomly selected. The pixel threshold of the diseased forest area is set. Each grid pixel in the target image is traversed. According to the given pixel threshold range of the diseased forest area, points that match the diseased pest area are selected. Isolated diseased pest area points are removed. The contour around the dense diseased pest area points is used as the second fitted contour map. The first fitted contour map is overlapped with the second fitted contour map, and the non-overlapping edges are processed by Gaussian blurring to generate a gradient mask to obtain the pest and disease area. The proportion of the pest and disease area in the forest area is calculated to obtain the pest and disease density. The pest and disease area and the pest and disease density are saved as a set of forest area pest and disease features.

[0009] Preferably, the processing logic for predicting the feature set of forest pests and diseases based on the forest pest and disease spread detection model is as follows: Based on the dynamic recording of forest pest and disease characteristics in time series, the area growth rate of pest and disease areas at adjacent time points is calculated. The coordinates of one edge point of the pest and disease area are randomly selected, and the displacement of the edge point coordinates at adjacent time points is recorded to obtain the displacement rate and displacement direction of the edge point coordinates per unit time. The area growth rate, displacement rate and displacement direction of the edge point coordinates are aligned with the environmental parameters and forest area information according to the same timestamp to obtain a spatiotemporal state sequence. The displacement direction of the edge point is discretized into D directional states, and the area growth rate is discretized into G change levels. The Cartesian product of the D directional states and G change levels is solved to obtain S joint states, which represent the pest and disease spread state space. Based on the displacement direction of the edge point and the area growth rate, the joint states are mapped to the spatiotemporal state sequence to obtain a state transition sequence. The transition frequency matrix (S*S) is initialized, with each element of the transition frequency matrix being 0. The joint state transition frequency in the state transition sequence is counted to obtain the initial state transition frequency matrix. The logic for processing the joint state transition frequency in the statistical state transition sequence is as follows: Iterate through each time step t in the state transition sequence, and set the joint state at the current time step t as follows: The state at the next moment is , where i, j=1,2,3,...,S, increment the element in the i-th row and j-th column of the transition frequency matrix by 1, and after traversing, obtain the initial state transition frequency matrix; Construct an environmental parameter likelihood function, traverse the spatiotemporal state sequence, statistically analyze the average area growth rate, extract the environmental parameters corresponding to the area growth rate being greater than the average, and calculate the average values ​​of temperature, humidity, soil moisture content, and soil pH in the environmental parameters respectively as the optimal environmental parameters. The optimal environmental parameters include optimal temperature, optimal humidity, optimal soil moisture content, and optimal soil pH. Save the optimal environmental parameters in the form of a vector as the optimal environmental parameter vector. Construct the environment likelihood function, the expression for which the environment likelihood function is calculated is: ; in, These are the environmental parameters corresponding to the spatiotemporal state sequence. For the optimal environmental parameter vector, Let k be the standard deviation of the k-th environmental parameter. It is the square of the difference between the current environmental parameters and the optimal environmental parameters. e is the natural constant. The environmental likelihood function value represents the likelihood between the current environmental parameters and the optimal environmental parameters. A pest and disease spread regulating factor is set, and the calculation expression of the pest and disease spread regulating factor is: ; in, The environmental likelihood function value. λ is the proportionality coefficient, e is the natural constant, and λ is the pest and disease spread regulation factor; The initial state transition frequency matrix is ​​corrected in real time using the pest and disease diffusion factor to obtain the time-varying transition matrix. The processing logic for real-time correction of the initial state transition frequency matrix using the pest and disease diffusion factor is as follows: If the area of ​​the pest and disease region in the next joint state is greater than the area of ​​the pest and disease region in the previous joint state, the pest and disease diffusion factor is multiplied by the initial state transition frequency matrix to obtain the first time-varying transition matrix. If the area of ​​the pest and disease region in the next combined state is smaller than the area of ​​the pest and disease region in the previous combined state, the reciprocal of the pest and disease diffusion factor is multiplied by the initial state transition frequency matrix to obtain the second time-varying transition matrix. If the area of ​​the pest and disease region in the next joint state is at the same level of change as the area of ​​the pest and disease region in the previous joint state, then the initial state transition frequency matrix is ​​used as the third time-varying transition matrix. Normalize each row of the first, second, and third time-varying transition matrices, and then calculate the time-varying transition matrix using Markov matrices. The current state transition sequence is multiplied by the time-varying transition matrix as a state vector to predict the area growth rate level and the edge point displacement direction as the diffusion state prediction value.

[0010] Preferably, the specific method for constructing a forestry pest and disease knowledge graph and detecting the disease probability of different tree species includes: The number of diseased trees is counted for each tree species. Based on historical disease records and combined with knowledge of forestry pests and diseases, a two-layer chain forestry pests and diseases knowledge graph is constructed. The establishment of the two-layer chain forestry pests and diseases knowledge graph includes: character annotation of the forestry pests and diseases knowledge based on deep learning named entity recognition technology; semantic association of forestry pests and diseases knowledge is linked through character annotation entity analysis to form a link classification; semantic association between entities is determined according to predefined character annotations of forestry pests and diseases knowledge; relevant information on forest disease occurrence is obtained according to the search content; and the establishment is completed when the forestry pests and diseases knowledge graph retrieves a new cause of disease. The system dynamically calculates the ratio of different diseased trees to the total number of diseased trees in the forest area per unit time, generating a dynamic tree species disease incidence ratio. Based on the dynamic tree species disease incidence ratio, a tree species disease incidence probability wave is generated. Based on the disease incidence probabilities of different tree species, an effective disease incidence probability threshold is set. The disease incidence ratio of tree species within the effective disease incidence probability threshold is extracted and its average value is calculated to obtain the current tree species disease incidence probability. A probabilistic graphical model is constructed using a Bayesian network, and tree species, environmental parameters, and forest area information are set as nodes to generate a conditional probability table. An inference library is called, and the data obtained by the ground sensors is connected to the probabilistic graphical model. The tree species disease incidence probability is calculated using variable elimination inference. The forest pest and disease severity level is defined based on the tree species disease incidence probability.

[0011] Preferably, the knowledge in the field of forestry diseases includes forest diseases and forest pests; The forest diseases mentioned include disease pathogens, disease symptoms, disease control, ecological environment, disease alternatives, affected parts, disease occurrence patterns, and control agents; The forest pests mentioned include pest categories, pest symptoms, pest control, alternative names for pests, affected parts of the body, natural enemies, and pesticides.

[0012] Preferably, the specific method for defining the severity level of forest pests and diseases based on the incidence probability of the tree species includes: Based on the calculated tree species disease incidence rate and forest area information, a judgment rule engine is set up. When the tree species disease incidence rate is within the first threshold range and there is no historical disease record, the first response measure is triggered. When the tree species disease incidence rate is within the second threshold range and the disease occurs in the neighboring area, the second response measure is triggered. When the tree species disease incidence rate is within the third threshold range and the environmental parameters are normal, the third response measure is triggered. The first threshold range is less than the second threshold range and the third threshold range is less than the third threshold range. The first response measures include routine drone monitoring; The second countermeasure includes locally spraying pesticides on tree species with a high probability of disease incidence around the affected area; The third response measure includes emergency isolation of the affected area and its surroundings, and the use of unmanned aerial vehicle (UAV) systems to control pests and diseases.

[0013] Preferably, step S5 specifically includes: The system monitors the disease types in disease-prone areas, retrieves forestry disease and pest knowledge graphs based on the monitored disease types, and reverse-engineers pathogens, disease pathogens, and ecological environment. Based on environmental parameter variables read from ground sensors, it uses causal discovery algorithms to generate a causal path map of forest disease occurrence related to the knowledge graph and visualizes the potential causes of diseases in forest areas.

[0014] Preferably, the specific method for generating a knowledge graph-related causal path map of forest disease incidence using a causal discovery algorithm includes: The environmental parameters are aligned in time sequence. Based on the knowledge graph of forestry pests and diseases, the pathogenic factors, pathogens and ecological environment are connected in pairs to construct a completely undirected graph. For each edge, it is tested whether there is a set of conditions that associates A with B with C, where A, B and C represent environmental parameters, tree species and pathogens. If a set of conditions exists, the link is retained; if no set of conditions exists, the link is deleted.

[0015] Secondly, a network security protection system based on big data includes a data acquisition module, a pest and disease area detection module, a condition prediction module, and a knowledge graph module. The data acquisition module is used to acquire target monitoring data sets for various blocks in the forest area, including target images, environmental parameters, and forest area information. The pest and disease area detection module is used to detect pest and disease areas in the target image, obtain a first fitted contour map based on the first detection, obtain a second fitted contour map based on the second detection, and superimpose and fuse the first fitted contour map and the second fitted contour map to obtain a set of forest area pest and disease features. The condition prediction module is used to construct a multi-task forest area detection model based on the set of forest area pest and disease characteristics. The knowledge graph module is used to construct a knowledge graph of forestry diseases and pests, detect the probability of disease occurrence in different types of trees, optimize the generation of forest disease and pest severity levels, and analyze potential causes of diseases in forest areas based on the forest disease and pest severity levels.

[0016] The beneficial effects of this invention are as follows: Through a two-stage contour fusion mechanism for detecting pest and disease areas, this invention achieves refined localization of lesion boundaries, significantly reducing the risk of false positives and false negatives. This invention does not rely on a single threshold or depth model for lesion identification; instead, it locates the starting and ending points of pests and diseases based on pixel gradient mutations. It employs a first fitted contour detection based on pixel grayscale adaptation and a second fitted contour detection based on lesion pixel threshold clustering. By overlapping and fusing the two fitted contours and applying Gaussian blur gradient processing to non-overlapping edge regions, the pest and disease area maintains both boundary accuracy and internal lesion integrity, thus avoiding edge breakage problems caused by changes in lighting, canopy shading, or noise interference in existing technologies. This invention no longer simply outputs the probability of pest and disease occurrence, but instead jointly discretizes the area change of the pest and disease region and the displacement direction of edge points to construct a pest and disease diffusion state space. Based on the state transition sequence, it statistically analyzes the state transition frequency to form an initial transition matrix. This allows the invention not only to determine whether pests and diseases have spread, but also to dynamically predict the area growth rate and displacement direction of pest and disease diffusion, significantly enhancing the value of the prediction results for prevention and control decisions. Addressing the problem that existing models treat environmental parameters as static feature inputs and fail to reflect the impact of continuous environmental changes, this invention quantifies the relationship between the current environment and the optimal diffusion cycle by constructing an environmental parameter likelihood function. The system determines the degree of matching between different environments and generates pest and disease spread regulation factors to correct the state transition matrix in real time. This allows the spread prediction results to be adjusted according to changes in temperature, humidity, and soil conditions, thereby avoiding distortion of prediction results in the event of sudden environmental changes. The invention also constructs a forestry pest and disease knowledge graph with a two-layer chain structure. It distinguishes and models knowledge in the fields of disease mechanism and tree species susceptibility with specific forest area instances. It also constructs clear causal relationship links through named entity recognition and semantic association. This enables the system to not only consider statistical features when calculating the probability of tree species disease, but also to distinguish related phenomena from the real causes, effectively reducing the probability of misjudgment. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the basic process of a forestry pest and disease prediction method based on big data, provided as an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the basic process of a forestry pest and disease prediction system based on big data, provided as an embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for predicting forestry pests and diseases based on big data is provided, comprising: Step S1: Obtain target monitoring data sets for each block in the forest area, wherein the target monitoring data sets include target images, environmental parameters, and forest area information; Step S2: Detect pest and disease areas in the target image to obtain a set of forest pest and disease features; Step S3: Based on the forest area diffusion detection model, predict the feature set of forest area pests and diseases to obtain the predicted value of diffusion status; Step S4: Construct a knowledge graph of forestry diseases and pests, detect the probability of disease occurrence in different types of trees, and optimize the generation of forest disease and pest severity levels; Step S5: Analyze the potential causes of diseases in the forest area based on the severity level of forest pests and diseases.

[0021] In this embodiment, by integrating multi-source data such as images captured by drones, IoT sensors, and forestry databases, and combining a multi-task model, the system generates the disease and pest area growth rate and the change rate of coordinate displacement of disease and pest edge points through time-series prediction. Then, by constructing a forestry disease and pest knowledge graph, the system digitizes knowledge such as tree species and environmental parameters to dynamically calculate the disease probability and generate a quantitative risk level. Finally, based on causal analysis, the system locates root causes such as soil pH and vector insects to generate targeted prevention and control plans, thereby improving the efficiency of prevention and control.

[0022] Step S1 specifically includes: Step S11: Divide the forest area into several blocks according to tree species, and use drone aerial photography technology to take several target images of several blocks. The target images are used to capture the current state of the trees and the status of pests and diseases. Step S12: Install ground sensors in several blocks to obtain environmental parameters of the forest area. The ground sensors include temperature and humidity sensors and soil monitoring sensors. The environmental parameters include temperature, humidity, soil moisture content and soil pH. Step S13: Obtain forest area information based on historical forest area records. The forest area information includes tree species, tree height, trunk diameter, tree age, and historical disease records. Step S14: Save the target image, environmental parameters, and forest area information into a target monitoring data set according to n blocks.

[0023] Based on the GIS map of the forest area and the distribution of tree species, the forest area was divided into several blocks using regular grid partitioning. Using a drone equipped with a multispectral lens, the flight altitude was set to 50 meters and the image overlap rate was greater than 70%. A systematic aerial photography of the forest area was carried out to obtain target images. A temperature and humidity sensor and a soil sensor were deployed at the center of each block. Data was transmitted wirelessly to the edge gateway and uploaded every hour. Forest area information was retrieved from the forestry database. The collected data was then timestamped and spatially unified using spatial block IDs. This embodiment significantly improves the accuracy and efficiency of forest pest and disease monitoring through multi-source data collaborative acquisition and intelligent processing. First, a block division strategy combining regular grids and adaptive clustering is adopted to ensure the scientific nature and representativeness of monitoring coverage. Combined with UAV multispectral aerial photography and ground sensor networks, integrated aerial and ground data acquisition is achieved. Second, the problem of real-time data transmission in remote forest areas is solved through LoRaWAN low-power transmission and edge computing preprocessing. Innovatively, historical disease records and real-time monitoring data are spatiotemporally aligned to construct a structured multimodal database. Compared with traditional methods, this solution upgrades single-point monitoring to full-domain perception and reduces the cost of manual inspection.

[0024] Specific methods for detecting pest and disease areas in target images include: Median filtering is used to remove noise and perform binary grayscale conversion on the target image, and the resolution is adjusted to W×H. The target image is divided into a grid. Starting from the top left corner of the target image, the pixel gradient magnitudes in the horizontal and vertical directions are calculated. The pixel gradient magnitudes within the grid are statistically analyzed, and the average gradient magnitude of each grid is calculated. The average gradient magnitudes are iterated, and the two adjacent grids with the largest differences are found as the starting and ending points of the pest and disease region. Pest and disease region detection is set up, including a first detection and a second detection. The coordinates of the starting and ending points are found on the target image, and the sampling interval is set. The first detection is used to obtain the first fitted contour map. The pixel grayscale values ​​of the target image are queried. Starting from the starting point, the sampling interval is set at a set unit angle. The process iterates through the pixel grayscale values ​​of the target image. When the pixel grayscale value matches the pixel fit of the diseased tree, the feature points of the diseased tree are extracted and the location of the feature points is marked until the coordinates of the last point are returned. The enclosed area is fitted into the first fitted contour map, and the marked points of the diseased tree area are generated. The proportion of the area of ​​the diseased tree area to the area of ​​the entire target image is calculated. The second detection is used to obtain the second fitted contour map. Pixels of the grid area on the target image are randomly selected. The pixel threshold of the diseased forest area is set. Each grid pixel in the target image is traversed. According to the given pixel threshold range of the diseased forest area, points that match the diseased area are selected. Isolated diseased area points are removed. The contour around the dense diseased area points is used as the second fitted contour map. The first fitted contour map is overlaid with the second fitted contour map. The non-overlapping edges are processed by Gaussian blurring to generate a gradient mask to obtain the disease and pest areas. The proportion of the disease and pest areas in the forest area is calculated to obtain the disease and pest density. The disease and pest areas and the disease and pest density are saved as a set of forest area disease and pest features.

[0025] Median filtering is used to remove noise, the image is converted to grayscale, and the image resolution W is uniformly adjusted. H=1024 1024. To avoid subsequent grid deviations, gradient magnitude analysis is performed. The Sobel operator is used to calculate the horizontal and vertical gradients. The average gradient magnitude of all pixels in each grid is calculated to reflect the edge strength of the area. All grids are traversed to find the two adjacent grids with the largest difference in gradient mean, which are used as the starting and ending points of the diseased area. Since the edge of the disease usually causes abrupt gradient changes, the two adjacent grids with the largest difference in gradient magnitude are selected as the boundary between the diseased area and the healthy tree. The first detection method is active contour fitting. Starting from the starting point, the unit step size is set to 1 degree, and the system scans outward in polar coordinates. The pixel fit of diseased trees is defined as 50-100. Pixels that meet the threshold are found along the scanning direction and marked as feature points. The scanning continues until the end point is returned. All feature points are connected to form the first fitted contour map. The area within the contour is calculated using Green's formula, and the proportion of the area to the total image area is used to obtain the lesion coverage probability. The second detection method is threshold clustering. Five pixels are randomly selected from the grid. The range of pixel thresholds for lesion forest areas is set to gray values ​​less than 100 and gradient values ​​greater than 30. Potential lesions are screened. The extracted potential lesions are denoised, and isolated points with fewer than 3 lesions in the neighborhood are removed. Convex hull calculation is performed on the dense point set to generate the second fitted contour map. The first and second fitted contour maps are superimposed, and Gaussian blur is applied to the non-overlapping edge areas to smooth the transition. The merged area is the accurate disease range, while preserving the accuracy of the boundary and the integrity of the internal lesions. The area of ​​the diseased and pest-infested areas of all blocks is counted, and the overall disease and pest density of the forest area is calculated. In this embodiment, a preprocessing method combining median filtering and binary grayscale conversion is used to effectively eliminate image noise and retain key edge features, providing a high-quality data foundation for subsequent analysis. Secondly, an innovative gridded analysis method based on gradient magnitude is proposed. By calculating the difference in the average gradient of adjacent grids, the boundary of the pest and disease area is automatically located, significantly improving the accuracy and efficiency of pest edge detection. A two-stage contour detection strategy is adopted. The first detection achieves accurate boundary fitting through pixel grayscale adaptation, and the second detection ensures that no small lesions are missed based on threshold screening. The two complement each other to improve the detection completeness rate and realize real-time identification and early warning of pests and diseases.

[0026] The processing logic for predicting the feature set of forest pests and diseases based on the forest pest and disease spread detection model is as follows: Based on the dynamic recording of forest pest and disease characteristics in time series, the area growth rate of pest and disease areas at adjacent time points is calculated. The coordinates of one edge point of the pest and disease area are randomly selected, and the displacement of the edge point coordinates at adjacent time points is recorded to obtain the displacement rate and displacement direction of the edge point coordinates per unit time. The area growth rate, displacement rate and displacement direction of the edge point coordinates are aligned with the environmental parameters and forest area information according to the same timestamp to obtain a spatiotemporal state sequence. The displacement direction of the edge point is discretized into D azimuth states, and the area growth rate is discretized into G change levels. A Cartesian product is then performed on the D azimuth states and G change levels to obtain S joint states, which represent the pest and disease spread state space. Based on the displacement direction of the edge points and the area growth rate, the joint state is mapped to the spatiotemporal state sequence to obtain the state transition sequence. The transition frequency matrix (S*S) is initialized, with each element of the transition frequency matrix being 0. The joint state transition frequency in the state transition sequence is counted to obtain the initial state transition frequency matrix. The logic for processing the joint state transition frequency in the statistical state transition sequence is as follows: Iterate through each time step t in the state transition sequence, and set the joint state at the current time step t as follows: The state at the next moment is , where i, j=1,2,3,...,S, increment the element in the i-th row and j-th column of the transition frequency matrix by 1, and after traversing, obtain the initial state transition frequency matrix; Construct an environmental parameter likelihood function, traverse the spatiotemporal state sequence, statistically analyze the average area growth rate, extract the environmental parameters corresponding to the area growth rate being greater than the average, and calculate the average values ​​of temperature, humidity, soil moisture content, and soil pH in the environmental parameters respectively as the optimal environmental parameters. The optimal environmental parameters include optimal temperature, optimal humidity, optimal soil moisture content, and optimal soil pH. Save the optimal environmental parameters in the form of a vector as the optimal environmental parameter vector. Construct the environment likelihood function, the expression for which the environment likelihood function is calculated is: ; in, These are the environmental parameters corresponding to the spatiotemporal state sequence. For the optimal environmental parameter vector, Let be the standard deviation of the k-th environmental parameter, representing the range of fluctuation of that environmental parameter or the sensitivity of pests and diseases to that parameter. The closer the value is to 1, the more suitable the environment is for the spread of pests and diseases. It is the square of the difference between the current environmental parameters and the optimal environmental parameters. e is the natural constant. The environmental likelihood function value represents the likelihood between the current environmental parameters and the optimal environmental parameters. A pest and disease spread regulating factor is set, and the calculation expression of the pest and disease spread regulating factor is: ;

[0027] in, The environmental likelihood function value. λ is the proportionality coefficient (set to 1), e is the natural constant, and λ is the pest and disease spread regulation factor. The initial state transition frequency matrix is ​​corrected in real time using the pest and disease diffusion factor to obtain the time-varying transition matrix. The processing logic for real-time correction of the initial state transition frequency matrix using the pest and disease diffusion factor is as follows: If the area of ​​the pest and disease region in the next joint state is greater than the area of ​​the pest and disease region in the previous joint state, the pest and disease diffusion factor is multiplied by the initial state transition frequency matrix to obtain the first time-varying transition matrix. If the area of ​​the pest and disease region in the next combined state is smaller than the area of ​​the pest and disease region in the previous combined state, the reciprocal of the pest and disease diffusion factor is multiplied by the initial state transition frequency matrix to obtain the second time-varying transition matrix. If the area of ​​the pest and disease region in the next joint state is at the same level of change as the area of ​​the pest and disease region in the previous joint state, then the initial state transition frequency matrix is ​​used as the third time-varying transition matrix. Normalize each row of the first, second, and third time-varying transition matrices, and then calculate the time-varying transition matrix using Markov matrices. The current state transition sequence is multiplied by the time-varying transition matrix as a state vector to predict the area growth rate level and the edge point displacement direction as the diffusion state prediction value.

[0028] In this embodiment, the area growth rate and edge point displacement are first calculated. For adjacent times t and t-1, the area growth rate of the pest-affected area is calculated. On the outline of the pest-affected area at time t-1, an edge point is randomly selected. On the outline at time t, the corresponding point of the edge point is determined by finding the nearest point. The displacement vector of the edge point and the corresponding point is calculated per unit time, thereby obtaining the displacement change rate and displacement direction. The area growth rate, edge point coordinate displacement change rate, and edge point displacement direction at each time t are aligned with environmental parameters and forest area information according to the same timestamp to obtain a spatiotemporal state sequence. The edge point displacement direction is discretized into D directional states. In this embodiment, D=8 (e.g., North, Northeast, East, Southeast, South, Southwest, West, Northwest). The area growth rate is discretized into G change levels. In this embodiment, G=3, which is equivalent to 3 change levels, namely -1, 0, and 1. Level -1 indicates that the area of ​​pest and disease spread has shrunk (area growth rate). -0.05), level 0 indicates stable pest and disease control (-0.05). Area growth rate 0.05), Level 1 indicates an increase in the area affected by pests and diseases (area growth rate). 0.05), performing a Cartesian product on D azimuth states and G change levels, yields =24 joint states, for example, joint Represented as: Northeast direction, level 1, 24 joint states constitute the pest and disease spread state space. Traversing the state transition sequence, the data at each time step is mapped to the corresponding joint state, forming a state sequence from which all adjacent time step state transition pairs are extracted. A 24×24 transition frequency matrix is ​​initialized with all elements set to 0. Traversing all state transition pairs, if a transition occurs from state... Transferred to Then, increment the element in the i-th row and j-th column of matrix F by 1. After traversing the matrix, normalize each row of matrix F (i.e., divide each row's element by the sum of its elements) to obtain the initial state transition probability matrix. This matrix represents the statistical law of the transition between pest and disease spread states under historical average conditions. From the state transition sequence, select all moments when the area growth rate G_t is greater than the historical average growth rate. Calculate the average values ​​of the environmental parameters (T, H, W, pH) corresponding to these moments to obtain the optimal environmental parameter vector, which represents the center point of the environmental conditions most favorable for pest and disease spread. For the current environmental parameter vector, calculate the likelihood between the current environmental parameter vector and the optimal environmental parameter vector. Use the modified multivariate Gaussian distribution density function as the environmental likelihood function. The obtained likelihood function value L is between 0 and 1. The closer L is to 1, the closer the current environment is to the historical optimal spread environment, and the more favorable it is for pest and disease spread; conversely, the less favorable it is. The environmental likelihood value L is transformed into the moderating strength of the state transition probability. A modified pest and disease spread moderating factor is designed. When L=1, the environment is optimal. =1, no adjustment; when L<1 When L < 1, it suppresses the transition probability; when L > 1, >1, which enhances the probability of disease and pest spread. This embodiment realizes quantitative prediction of the coupling between the disease and pest spread trend and the external environment by state-based analysis of the disease and pest spread process and dynamic modification of the initial state transition probability matrix using the environmental likelihood function. It obtains the predicted value of the spread state at the next moment, effectively mitigating the influence of environmental fluctuations on historical disease and pest spread patterns.

[0029] Specific methods for constructing a knowledge graph of forestry pests and diseases to detect the probability of disease occurrence in different tree species include: The number of diseased trees is counted according to tree species. Based on historical disease records and combined with knowledge of forestry pests and diseases, a two-layer chain forestry pests and diseases knowledge graph is constructed. The establishment of the two-layer chain forestry pests and diseases knowledge graph includes: character annotation of forestry pests and diseases knowledge based on deep learning named entity recognition technology, semantic association analysis of forestry pests and diseases knowledge through character annotation entity analysis to form a chain classification, determination of semantic association between entities based on predefined character annotations of forestry pests and diseases knowledge, obtaining relevant information on forest disease occurrence based on search content, and completion of the establishment when a new cause of disease is found in the forestry pests and diseases knowledge graph. The system dynamically calculates the ratio of different diseased trees to the total number of diseased trees in the forest area per unit time, generating a dynamic tree species disease incidence ratio. Based on the dynamic tree species disease incidence ratio, it generates a tree species disease incidence probability wave. Based on the disease incidence probabilities of different tree species, it sets an effective disease incidence probability threshold, extracts the disease incidence ratio of tree species within the effective disease incidence probability threshold, calculates the average value, and obtains the current tree species disease incidence probability. It uses a Bayesian network to construct a probabilistic graphical model, setting tree species, environmental parameters, and forest area information as nodes to generate a conditional probability table. It calls the inference library and simultaneously integrates data acquired by ground sensors into the probabilistic graphical model, and uses variable elimination inference to calculate the tree species disease incidence probability. Based on the tree species disease incidence probability, it defines the forest pest and disease severity level.

[0030] Historical disease data, tree species, and tree age were extracted from forestry information. Based on forestry research papers and disease control manuals obtained online, entities were labeled using a BERT-based model. An entity instance was defined as: "Pine wilt disease (disease) is caused by pine wilt nematode (pathogen), the susceptible tree species is pine (tree species), and high temperature (environmental conditions) accelerates its spread." Relationships between entities were extracted. The first layer represents static knowledge, with nodes set as tree species, disease, and environmental thresholds, and edges set as susceptibility, pathogenic factors, and inhibitory conditions. The second layer represents dynamic instances, with nodes set as instances of diseased trees in the current forest area. (Pine trees in area A-001), with edges set to located, infected, and adjacent transmission. The above two layers are stored as a forestry pest and disease knowledge graph. Dynamic tree species disease probability calculation: First, calculate the disease incidence rate of each tree species per unit time. For each tree species, calculate the standard deviation of the disease incidence rate over time. If the current rate is greater than the peak threshold, delete the rate. If it is between the peak and trough, retain the rate. In this embodiment, a two-layer knowledge graph is used to achieve the linkage of "theoretical knowledge + real-time examples". The Bayesian network is dynamically updated to adapt to seasonal or sudden disease changes. When the disease incidence rate jumps, response measures are automatically triggered.

[0031] Knowledge in the field of forestry diseases includes forest diseases and forest pests; Forest diseases include disease pathogens, disease symptoms, disease control, ecological environment, disease alternatives, affected parts, disease development patterns, and control agents; Forest pests include pest categories, pest symptoms, pest control, alternative names for pests, affected parts of the body, natural enemies, and pesticides.

[0032] In this embodiment, diseases (eight categories of elements such as pathogens, symptoms, and control) are structurally linked with pests (seven categories such as categories, natural enemies, and pesticides) to form the most comprehensive knowledge base in the forestry field, supporting multi-angle etiological tracing (such as locking the pathogen type by “affected parts + disease pattern”).

[0033] Specific methods for defining the severity level of forest pests and diseases based on the probability of disease occurrence by tree species include: Based on the calculated tree species disease probability and forest area information, a judgment rule engine is set up. When the tree species disease probability is within the first threshold range and there is no historical disease record, the first response measure is triggered. When the tree species disease probability is within the second threshold range and the disease occurs in the neighboring area, the second response measure is triggered. When the tree species disease probability is within the third threshold range and the environmental parameters are normal, the third response measure is triggered. The first threshold range is less than the second threshold range and less than the third threshold range. The first response measures include routine drone monitoring; The second response measure includes local spraying of pesticides on tree species with a high probability of disease incidence around the affected area. The third response measures include emergency isolation of the affected areas and surrounding areas, and the use of drone systems to control pests and diseases.

[0034] In this embodiment, the first threshold is set to 0-0.3, the second threshold is set to 0.3-0.7, and the third threshold is set to 0.7-1. The risk level is divided according to the three dimensions of disease incidence probability, historical record, and spatial correlation to avoid misjudgment caused by traditional single threshold. For example, when the disease incidence probability of pine trees reaches 65% (level 2) and there are cases in the neighboring area, local spraying is automatically triggered instead of spraying the entire area, reducing pesticide waste and reducing timber loss caused by disease.

[0035] Step S5 specifically includes: The system monitors the disease types in disease-prone areas, retrieves forestry disease and pest knowledge graphs based on the monitored disease types, and reverse-engineers pathogens, disease pathogens, and ecological environment. Based on environmental parameter variables read from ground sensors, it uses causal discovery algorithms to generate a causal path map of forest disease occurrence related to the knowledge graph and visualizes the potential causes of diseases in forest areas.

[0036] In this embodiment, based on the detected disease type, the system automatically associates pathogenic factors (such as fungal pathogens, sensitive ecological environments, etc.) with knowledge graphs, overcoming the limitations of traditional methods that rely on human experience. For example, if the system finds that "pine wilt disease" requires vector insects for transmission, and the sensor detects that the current temperature is 28℃ (suitable for insect activity), it directly identifies "high temperature + vector insects" as the key cause. Secondly, it uses the analysis of environmental parameters (temperature / humidity / soil pH) and the causal relationship between the disease occurrence (such as "soil acidification → tree resistance ↓ → disease susceptibility" rather than just statistical correlation) to avoid misjudgment. When the sensor detects a sudden increase in humidity (such as continuous rainfall), it automatically updates the causal graph to "rainfall → humidity ↑ → accelerated spore reproduction".

[0037] Specific methods for generating knowledge graph-related causal path maps of forest disease incidence using causal discovery algorithms include: Environmental parameters are aligned in time sequence. Based on the knowledge graph of forestry pests and diseases, pathogens, and ecological environment are connected in pairs to construct a completely undirected graph. For each edge, it is tested whether there is a set of conditions that associates A with B with C, where A, B, and C represent environmental parameters, tree species, and pathogens. If a set of conditions exists, the link is retained; otherwise, the link is deleted.

[0038] In this embodiment, by testing the conditional dependencies between environmental parameters (A), tree species (B), and pathogens (C) (such as whether "humidity → disease probability" remains significant after temperature control), the algorithm effectively distinguishes between true causality (such as soil pH directly affecting resistance) and spurious correlation (such as humidity and disease occurring simultaneously but without direct causation). The algorithm finds that "temperature and pine wilt disease" loses statistical significance after controlling the activity of vector insects, thus excluding temperature as the direct cause of disease and locating the true causal chain of "temperature → insect reproduction → disease spread". This embodiment combines data-driven causal discovery with domain knowledge verification, achieving a leap from "empirical guessing" to "scientific attribution".

[0039] Example 2, refer to Figure 2 In another embodiment of the present invention, which differs from the first embodiment, a method for predicting forestry pests and diseases based on big data is provided. To verify and illustrate the technical effects of the method, this embodiment compares the traditional technical solution with the method of the present invention, and compares the experimental results with scientific demonstration methods to verify the real effect of the method.

[0040] A forestry pest and disease prediction system based on big data includes a data acquisition module, a pest and disease area detection module, a spread prediction module, a knowledge graph module, and a pathogenesis analysis module. The data acquisition module is used to acquire target monitoring data sets for various blocks in the forest area. The target monitoring data sets include target images, environmental parameters, and forest area information. The pest and disease area detection module is used to detect pest and disease areas in the target image. It obtains a first fitted contour map based on the first detection and a second fitted contour map based on the second detection. The first fitted contour map and the second fitted contour map are superimposed and fused to obtain a set of forest area pest and disease features. The diffusion prediction module is used to predict the feature set of forest pests and diseases based on the forest diffusion detection model and obtain the predicted value of diffusion status. The knowledge graph module is used to construct a knowledge graph of forestry diseases and pests, detect the probability of disease occurrence in different types of trees, and optimize the generation of forest disease and pest severity levels. The etiology analysis module is used to analyze potential causes of diseases in forest areas based on the severity level of forest pests and diseases.

[0041] In this embodiment, the data acquisition module combines multi-source data, including UAV imagery (target images), ground sensors (environmental parameters), and forestry databases (forest area information), to construct a three-dimensional monitoring information network, solving the blind spot problem of traditional manual inspections. The pest and disease area detection module integrates contour fitting complementarity. The first detection (based on pixel grayscale adaptation) accurately locates the disease boundary, and the second detection (threshold clustering) captures internal lesions. After fusion, the disease detection rate is improved. The multi-task model simultaneously predicts the disease type (classification task), spread rate (regression task), and risk area (segmentation task), outputting multi-dimensional features of the disease. Finally, the knowledge graph module retrieves knowledge graphs in reverse according to the disease level, generating interpretable paths and root causes of disease, improving targeted forest area management and reducing forestry economic losses.

[0042] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

Claims

1. A method for predicting forestry pests and diseases based on big data, characterized in that, include: Step S1: Obtain target monitoring data sets for each block in the forest area, wherein the target monitoring data sets include target images, environmental parameters, and forest area information; Step S2: Detect pest and disease areas in the target image to obtain a set of forest pest and disease features; Step S3: Based on the forest area diffusion detection model, predict the feature set of forest area pests and diseases to obtain the predicted value of diffusion status; Step S4: Construct a knowledge graph of forestry diseases and pests, detect the probability of disease occurrence in different types of trees, and optimize the generation of forest disease and pest severity levels; Step S5: Analyze the potential causes of diseases in the forest area based on the severity level of forest pests and diseases.

2. The forestry pest and disease prediction method based on big data as described in claim 1, characterized in that: Step S1 specifically includes: Step S11: Divide the various blocks in the forest area into several blocks according to tree species, and use drone aerial photography technology to take several target images of several blocks. The target images are used to capture the current state of the trees and the state of pests and diseases. Step S12: Install ground sensors in several blocks to obtain environmental parameters of the forest area. The ground sensors include temperature and humidity sensors and soil monitoring sensors. The environmental parameters include temperature, humidity, soil moisture content and soil pH. Step S13: Obtain forest area information based on historical forest area records. The forest area information includes tree species, tree height, trunk diameter, tree age, and historical disease records. Step S14: Save the target image, environmental parameters, and forest area information into a target monitoring data set according to n blocks.

3. The forestry pest and disease prediction method based on big data as described in claim 2, characterized in that: The specific method for detecting pest and disease areas in the target image includes: The target image is noise-reducing and binary grayscale conversion is performed using median filtering, and the resolution is adjusted to W×H. The target image is divided into a grid. Starting from the top left corner of the target image, the pixel gradient magnitudes in the horizontal and vertical directions are calculated. The pixel gradient magnitudes within the grid are statistically analyzed, and the average gradient magnitude of each grid is calculated. The average gradient magnitudes are iterated, and the two adjacent grids with the largest differences are found as the starting and ending points of the pest and disease region. Pest and disease region detection is set up, which includes a first detection and a second detection. The coordinates of the starting and ending points are found on the target image, and a sampling interval is set. The first detection is used to obtain a first fitted contour map. The pixel grayscale values ​​of the target image are queried, starting from the starting point position, to set... The pixel grayscale values ​​of the target image are traversed at a fixed unit angle. When the pixel grayscale value matches the pixel fit of the diseased tree, the feature points of the diseased tree are extracted and the location of the feature points is marked until the coordinates of the last point are returned. The enclosed area is fitted into a first fitted contour map, and the marked points of the diseased tree area are generated. The proportion of the area of ​​the diseased tree area to the area of ​​the entire target image is calculated. The second detection is used to obtain the second fitted contour map. Pixels of the grid area on the target image are randomly selected. The pixel threshold of the diseased forest area is set. Each grid pixel in the target image is traversed. According to the given pixel threshold range of the diseased forest area, points that match the diseased pest area are selected. Isolated diseased pest area points are removed. The contour around the dense diseased pest area points is used as the second fitted contour map. The first fitted contour map is overlapped with the second fitted contour map, and the non-overlapping edges are processed by Gaussian blurring to generate a gradient mask to obtain the pest and disease area. The proportion of the pest and disease area in the forest area is calculated to obtain the pest and disease density. The pest and disease area and the pest and disease density are saved as a set of forest area pest and disease features.

4. The forestry pest and disease prediction method based on big data as described in claim 3, characterized in that: The processing logic for predicting the feature set of forest pests and diseases based on the forest pest and disease spread detection model is as follows: Based on the dynamic recording of forest pest and disease characteristics in time series, the area growth rate of pest and disease areas at adjacent time points is calculated. The coordinates of one edge point of the pest and disease area are randomly selected, and the displacement of the edge point coordinates at adjacent time points is recorded to obtain the displacement rate and displacement direction of the edge point coordinates per unit time. The area growth rate, displacement rate and displacement direction of the edge point coordinates are aligned with the environmental parameters and forest area information according to the same timestamp to obtain a spatiotemporal state sequence. The displacement direction of the edge point is discretized into D directional states, and the area growth rate is discretized into G change levels. The Cartesian product of the D directional states and G change levels is solved to obtain S joint states, which represent the pest and disease spread state space. Based on the displacement direction of the edge point and the area growth rate, the joint states are mapped to the spatiotemporal state sequence to obtain a state transition sequence. The transition frequency matrix (S*S) is initialized, with each element of the transition frequency matrix being 0. The joint state transition frequency in the state transition sequence is counted to obtain the initial state transition frequency matrix. The logic for processing the joint state transition frequency in the statistical state transition sequence is as follows: Iterate through each time step t in the state transition sequence, and set the joint state at the current time step t as follows: The state at the next moment is , where i, j=1,2,3,...,S, increment the element in the i-th row and j-th column of the transition frequency matrix by 1, and after traversing, obtain the initial state transition frequency matrix; Construct an environmental parameter likelihood function, traverse the spatiotemporal state sequence, statistically analyze the average area growth rate, extract the environmental parameters corresponding to the area growth rate being greater than the average, and calculate the average values ​​of temperature, humidity, soil moisture content, and soil pH in the environmental parameters respectively as the optimal environmental parameters. The optimal environmental parameters include optimal temperature, optimal humidity, optimal soil moisture content, and optimal soil pH. Save the optimal environmental parameters in the form of a vector as the optimal environmental parameter vector. Construct the environment likelihood function, the expression for which the environment likelihood function is calculated is: ； in, These are the environmental parameters corresponding to the spatiotemporal state sequence. For the optimal environmental parameter vector, Let k be the standard deviation of the k-th environmental parameter. It is the square of the difference between the current environmental parameters and the optimal environmental parameters. e is the natural constant. The environmental likelihood function value represents the likelihood between the current environmental parameters and the optimal environmental parameters. A pest and disease spread regulating factor is set, and the calculation expression of the pest and disease spread regulating factor is: ； in, The environmental likelihood function value. λ is the proportionality coefficient, e is the natural constant, and λ is the pest and disease spread regulation factor; The initial state transition frequency matrix is ​​corrected in real time using the pest and disease diffusion factor to obtain the time-varying transition matrix. The processing logic for real-time correction of the initial state transition frequency matrix using the pest and disease diffusion factor is as follows: If the area of ​​the pest and disease region in the next combined state is greater than the area of ​​the pest and disease region in the previous combined state, the pest and disease diffusion factor is multiplied by the initial state transition frequency matrix to obtain the first time-varying transition matrix. If the area of ​​the pest and disease region in the next combined state is smaller than the area of ​​the pest and disease region in the previous combined state, the reciprocal of the pest and disease diffusion factor is multiplied by the initial state transition frequency matrix to obtain the second time-varying transition matrix. If the area of ​​the pest and disease region in the next joint state is at the same level of change as the area of ​​the pest and disease region in the previous joint state, then the initial state transition frequency matrix is ​​used as the third time-varying transition matrix. Normalize each row of the first, second, and third time-varying transition matrices, and then calculate the time-varying transition matrix using Markov matrices. The current state transition sequence is multiplied by the time-varying transition matrix as a state vector to predict the area growth rate level and the edge point displacement direction as the diffusion state prediction value.

5. The forestry pest and disease prediction method based on big data as described in claim 4, characterized in that: The specific methods for constructing a knowledge graph of forestry pests and diseases to detect the probability of disease occurrence in different tree species include: The number of diseased trees is counted for each tree species. Based on historical disease records and combined with knowledge of forestry pests and diseases, a two-layer chain forestry pests and diseases knowledge graph is constructed. The establishment of the two-layer chain forestry pests and diseases knowledge graph includes: character annotation of the forestry pests and diseases knowledge based on deep learning named entity recognition technology; semantic association of forestry pests and diseases knowledge is linked through character annotation entity analysis to form a link classification; semantic association between entities is determined according to predefined character annotations of forestry pests and diseases knowledge; relevant information on forest disease occurrence is obtained according to the search content; and the establishment is completed when the forestry pests and diseases knowledge graph retrieves a new cause of disease. The system dynamically calculates the ratio of different diseased trees to the total number of diseased trees in the forest area per unit time, generating a dynamic tree species disease incidence ratio. Based on the dynamic tree species disease incidence ratio, a tree species disease incidence probability wave is generated. Based on the disease incidence probabilities of different tree species, an effective disease incidence probability threshold is set. The disease incidence ratio of tree species within the effective disease incidence probability threshold is extracted and its average value is calculated to obtain the current tree species disease incidence probability. A probabilistic graphical model is constructed using a Bayesian network, and tree species, environmental parameters, and forest area information are set as nodes to generate a conditional probability table. An inference library is called, and the data obtained by the ground sensors is connected to the probabilistic graphical model. The tree species disease incidence probability is calculated using variable elimination inference. The forest pest and disease severity level is defined based on the tree species disease incidence probability.

6. The forestry pest and disease prediction method based on big data as described in claim 5, characterized in that: The knowledge in the field of forestry diseases includes forest diseases and forest pests; The forest diseases mentioned include disease pathogens, disease symptoms, disease control, ecological environment, disease alternatives, affected parts, disease occurrence patterns, and control agents; The forest pests mentioned include pest categories, pest symptoms, pest control, alternative names for pests, affected parts of the body, natural enemies, and pesticides.

7. The forestry pest and disease prediction method based on big data as described in claim 6, characterized in that: Specific methods for defining the severity level of forest pests and diseases based on the incidence probability of the tree species include: Based on the calculated tree species disease incidence rate and forest area information, a judgment rule engine is set up. When the tree species disease incidence rate is within the first threshold range and there is no historical disease record, the first response measure is triggered. When the tree species disease incidence rate is within the second threshold range and the disease occurs in the neighboring area, the second response measure is triggered. When the tree species disease incidence rate is within the third threshold range and the environmental parameters are normal, the third response measure is triggered. The first threshold range is less than the second threshold range and the third threshold range is less than the third threshold range. The first response measures include routine drone monitoring; The second countermeasure includes locally spraying pesticides on tree species with a high probability of disease incidence around the affected area; The third response measure includes emergency isolation of the affected area and its surroundings, and the use of unmanned aerial vehicle (UAV) systems to control pests and diseases.

8. The forestry pest and disease prediction method based on big data as described in claim 7, characterized in that: Step S5 specifically includes: The system monitors the disease types in disease-prone areas, retrieves forestry disease and pest knowledge graphs based on the monitored disease types, and reverse-engineers pathogens, disease pathogens, and ecological environment. Based on environmental parameter variables read from ground sensors, it uses causal discovery algorithms to generate a causal path map of forest disease occurrence related to the knowledge graph and visualizes the potential causes of diseases in forest areas.

9. The forestry pest and disease prediction method based on big data as described in claim 8, characterized in that: Specific methods for generating knowledge graph-related causal path maps of forest disease incidence using causal discovery algorithms include: The environmental parameters are aligned in time sequence. Based on the knowledge graph of forestry pests and diseases, the pathogenic factors, pathogens and ecological environment are connected in pairs to construct a completely undirected graph. For each edge, it is tested whether there is a set of conditions that associates A with B with C, where A, B and C represent environmental parameters, tree species and pathogens. If a set of conditions exists, the link is retained; if no set of conditions exists, the link is deleted.

10. A forestry pest and disease prediction system based on big data, implemented based on the forestry pest and disease prediction method based on big data as described in any one of claims 1-9, characterized in that: It includes a data acquisition module, a pest and disease area detection module, a condition prediction module, and a knowledge graph module; The data acquisition module is used to acquire target monitoring data sets for various blocks in the forest area, including target images, environmental parameters, and forest area information. The pest and disease area detection module is used to detect pest and disease areas in the target image, obtain a first fitted contour map based on the first detection, obtain a second fitted contour map based on the second detection, and superimpose and fuse the first fitted contour map and the second fitted contour map to obtain a set of forest area pest and disease features. The condition prediction module is used to predict the feature set of forest pests and diseases based on the forest area diffusion detection model to obtain the diffusion status prediction value. The knowledge graph module is used to construct a knowledge graph of forestry diseases and pests, detect the probability of disease occurrence in different types of trees, optimize the generation of forest disease and pest severity levels, and analyze potential causes of diseases in forest areas based on the forest disease and pest severity levels.

Images