Intelligent diagnosis system for animals and plants based on multi-source data

Abstract

The application relates to the technical field of intelligent diagnosis and recognition, in particular to an intelligent diagnosis system for animals and plants based on multi-source data; a geographical space risk atlas is recognized, and the historical state and intervention record of crop plants are obtained; remote sensing technology is used to collect data of the crop plants, multi-source remote sensing time series data are obtained based on a multispectral sensor, a hyperspectral sensor, a thermal infrared sensor and a laser radar sensor; the geographical space risk atlas, the multi-source remote sensing time series data and the historical state and intervention record are recognized, and risk abnormal plants are determined; the sound detection equipment and the ring electrode array are used to recognize the risk abnormal plants, sound characteristic data and a three-dimensional electrical characteristic space distribution map are obtained, and the abnormal state index of the crop plants is further recognized; the abnormal state index and three-dimensional coordinates of the crop plants are used for cluster analysis, the infection center and the infection area are determined, and an abnormal early warning is performed.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnostic and identification technology, specifically to an intelligent diagnostic system for plants and animals based on multi-source data. Background Technology

[0002] In existing technologies, computer vision and image analysis methods have been preliminarily applied to plant and animal health monitoring. For example, by acquiring visible light or multispectral images of crop plants, traditional image processing and machine learning algorithms are used to extract and classify visual phenotypic features such as the color and texture of leaf lesions to identify specific diseases. Some solutions also introduce sensors such as lidar to construct three-dimensional point cloud models of crop plants, and assess their growth status by analyzing spatial structural features such as canopy density. These methods have a certain role in identifying diseases that have already exhibited significant macroscopic characteristics.

[0003] However, existing analytical techniques have significant limitations. First, their identification targets are limited to the external optical or three-dimensional structural features of organisms. When lesions occur internally or in their early latent stages, the visible light image information may not have changed significantly, making it difficult for traditional feature engineering and pattern recognition algorithms to effectively detect and classify them. Second, different causes may lead to similar external visual patterns. This phenomenon of different diseases presenting with similar symptoms makes it easy to confuse relying solely on feature extraction from a single image modality, resulting in insufficient accuracy in scene understanding and decision-making.

[0004] Furthermore, while existing technologies have explored the application of data from different sensors, most remain at the level of independent data analysis or simple overlay of results, lacking a computational framework capable of deep feature-level fusion and collaborative analysis of external remote sensing images with other heterogeneous non-image data, such as physiological and biochemical data and environmental parameters.

[0005] To address this, an intelligent diagnostic system for plants and animals based on multi-source data is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent diagnostic system for plants and animals based on multi-source data, which uses cluster analysis of crop plant abnormality index and three-dimensional coordinates to determine infection centers and infection areas, and to provide early warning of abnormalities.

[0007] To achieve the above objectives, the present invention provides an intelligent diagnostic system for plants and animals based on multi-source data, comprising:

[0008] The data acquisition module obtains a geospatial risk map based on rootstock and soil parameters; it also obtains historical status and intervention records of crop plants, including historical status, intervention time, intervention measures, and intervention effects.

[0009] The feature recognition module uses remote sensing technology to periodically collect data on crop plants and obtain multi-source remote sensing time-series data; it constructs a deep risk recognition model to identify geospatial risk maps, multi-source remote sensing time-series data, and historical status and intervention records to determine plants with abnormal risks.

[0010] The infection assessment module collects data from at-risk and abnormal plants using acoustic detection equipment to obtain acoustic feature data; identifies at-risk and abnormal plants using a ring electrode array to generate a three-dimensional spatial distribution map of the root system's electrical properties; and identifies the abnormal state index of the crop plant based on the acoustic feature data and the three-dimensional spatial distribution map of the electrical properties.

[0011] The anomaly warning module uses cluster analysis based on the abnormal state index and three-dimensional coordinates of crop plants to determine the infection center and infection area, and then issues anomaly warnings.

[0012] The identification process of the geospatial risk map includes:

[0013] Obtain digital information maps and mark the rootstock parameters of crop plants and the soil parameters of the plots they are located on;

[0014] Based on the plant physiological characteristics database, a standardized rootstock resistance score was assigned to each rootstock, and a standardized soil susceptibility score was assigned to plots with different soil parameters according to the specific parameters of clay and sandy soil.

[0015] By using a weighted fusion algorithm, the rootstock resistance score and soil susceptibility score on the digital information map are calculated to generate a geospatial risk map.

[0016] The acquisition methods for the multi-source remote sensing time series data include: using a drone equipped with a multispectral sensor, a hyperspectral sensor, a thermal infrared sensor, and a lidar sensor to perform periodic remote sensing scans of crop plants;

[0017] Among them, vegetation index, chlorophyll content, and nitrogen status parameters are obtained through multispectral and hyperspectral sensors; canopy temperature is obtained through thermal infrared sensors to assess water stress caused by root damage; and canopy three-dimensional point cloud data is obtained through lidar sensors to calculate leaf area index and canopy density physical structure parameters.

[0018] The multi-dimensional data obtained in each acquisition cycle are used as time slices, and multi-source remote sensing time series data are formed according to the acquisition order.

[0019] The identification process of the deep risk identification model includes: using multi-source remote sensing time series data as the core, identifying crop plants with continuously declining health indicators through time series analysis algorithms, and obtaining dynamic decline characteristics.

[0020] Historical status and intervention records of crop plants are retrieved. Causal correlation analysis is performed based on historical status, intervention time, intervention measures, and intervention effects to obtain historical status correlation coefficients. Geospatial risk maps are used to obtain geospatial risk coefficients of crop plants. A comprehensive analysis is performed based on dynamic decline characteristics, historical status correlation coefficients, and geospatial risk coefficients to obtain a comprehensive risk score. Plants with abnormal risk are identified based on the comprehensive risk score.

[0021] The process of obtaining the historical state correlation coefficient includes: obtaining historical state and intervention records to obtain the historical state of crop plants, intervention time, intervention measures and intervention effects; based on the historical state, intervention time and intervention measures, querying the built-in agricultural knowledge base to dynamically generate a quantitative standard expected effect; comparing the intervention effect in the historical state and intervention records with the standard expected effect to calculate the effect difference score.

[0022] The historical status records are compared with the database of typical symptoms of phylloxera, and the symptom similarity score is calculated.

[0023] By using a preset fusion function, the effect difference score and symptom similarity score are combined and calculated to obtain the historical state correlation coefficient.

[0024] The process of generating a three-dimensional electrical property spatial distribution map of the root system includes: deploying a ring electrode array in the soil at the base of the plant at risk and abnormality, applying weak, multi-frequency high-frequency alternating current sequentially between different electrode pairs through the host system, and simultaneously measuring the voltage distribution on all other electrodes to collect and form a complete electrical response dataset.

[0025] The electrical response dataset is solved by inverse operation using a reconstruction algorithm to generate a three-dimensional matrix in units of voxels. Based on the three-dimensional matrix, a three-dimensional spatial distribution map of electrical characteristics is obtained, including the three-dimensional coordinates, impedance value, and capacitance characteristic parameters obtained by multi-frequency scanning for each voxel.

[0026] The process of identifying the abnormal state index of crop plants includes: monitoring the elastic waves generated in the xylem vessels of crop plants under water stress due to the rupture of the water column, and obtaining acoustic characteristic data;

[0027] The acoustic feature data is matched with the pre-set acoustic spectrum library of phylloxera threat to obtain acoustic correlation feature data. Based on the acoustic correlation feature data and the three-dimensional coordinates, impedance values ​​and capacitance parameters of voxels in the three-dimensional electrical characteristic spatial distribution map, the crop plants infected by phylloxera are identified.

[0028] By processing the spatial distribution map of three-dimensional electrical properties using a three-dimensional image segmentation algorithm, and based on the impedance value and the three-dimensional morphology of the diseased tissue, the total volume of the diseased tissue and the total volume of the healthy root system are identified and calculated, and the abnormal state index is obtained.

[0029] The process of identifying infection centers and infection areas includes: binding the abnormal state index of crop plants with three-dimensional geographic coordinates to form a spatial point dataset; using a density-based spatial clustering algorithm to analyze the spatial point dataset and identify crop plants that are spatially adjacent and whose infection indices all reach a preset high-risk threshold as a cluster;

[0030] The geometric center of the cluster with the highest density is determined as the infection center, and the convex hull and contour boundary of the geographical area covered by the cluster at the infection center are used to determine the infection area.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] 1. This invention integrates plant genetic resistance and soil environmental data to construct a quantitative and visualized geospatial risk map; it enables accurate prediction and spatial location of potential disease risks, transforming vague agricultural experience into a standardized data model. Compared to traditional extensive management relying on manual inspections, this method can scientifically guide the optimal allocation of monitoring resources, prioritizing attention to high-risk areas, significantly improving the probability of early disease detection and management efficiency; it not only provides a key static risk benchmark for subsequent dynamic monitoring but also provides a long-term decision-making basis for planting planning and soil improvement, reducing the possibility of large-scale disease outbreaks from the source.

[0033] 2. This invention constructs a deep identification model that integrates dynamic time-series data, historical records, and static spatial risks. Through time-series analysis algorithms, it accurately captures the continuous decline trajectory of crop plant health status. By cross-validating with historical records and geographical risks, it achieves in-depth analysis and confirmation of abnormal signals, significantly improving the accuracy and reliability of risk identification and effectively avoiding misjudgments that may be caused by a single data source.

[0034] 3. This invention achieves in-depth mining and intelligent analysis of crop plant historical health records through dynamic querying of the built-in agricultural knowledge base; it transforms qualitative historical descriptions into quantitative correlation coefficients with diagnostic value, solving the problem that historical data is difficult to directly utilize by the model; by comparing the actual intervention effect with the standard expected effect, it can accurately identify crop plants that respond abnormally to routine management measures, revealing the underlying causes that may exist beneath the surface symptoms; combined with similarity analysis with typical disease symptoms, it provides strong historical evidence support for early warning of root diseases, significantly enhancing the logical reasoning ability and accuracy of the diagnostic model.

[0035] 4. This invention creates a high-confidence method for diagnosing root diseases by synergistically analyzing acoustic detection and three-dimensional electrical characteristic spatial distribution maps. The matching of acoustic signals provides physiological functional evidence for abnormal areas found in the three-dimensional electrical characteristic spatial distribution maps, significantly improving the specificity and accuracy of diagnosis and effectively avoiding misdiagnosis. Furthermore, through intelligent segmentation and calculation of three-dimensional images, precise and non-destructive quantification of the degree of root infection is achieved. This provides a scientific basis for assessing the severity of disease in individual crop plants and also provides a unified data foundation for subsequent regional disease assessment and prevention and control decisions. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the structure of the intelligent plant and animal diagnostic system based on multi-source data of the present invention;

[0037] Figure 2 This is a schematic diagram of the data flow of the intelligent plant and animal diagnostic system based on multi-source data of the present invention;

[0038] Figure 3 This is a schematic diagram of the historical state correlation coefficient acquisition process of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example 1:

[0041] See Figure 1 and Figure 2 This invention proposes an intelligent diagnostic system for plants and animals based on multi-source data, wherein... Figure 1 This is a schematic diagram of the system structure. Figure 2 This is a schematic diagram of the system's data flow; further, the system includes:

[0042] The data acquisition module identifies geospatial risk maps based on rootstock and soil parameters; it also acquires historical status and intervention records of crop plants, obtaining information on the historical status of crop plants, intervention time, intervention measures, and intervention effects.

[0043] The identification process of the geospatial risk map includes:

[0044] Obtain digital information maps and mark the rootstock parameters of crop plants and the soil parameters of the plots they are located on;

[0045] Based on the plant physiological characteristics database, a standardized rootstock resistance score was assigned to each rootstock, and a standardized soil susceptibility score was assigned to plots with different soil parameters according to the specific parameters of clay and sandy soil.

[0046] By using a weighted fusion algorithm, the rootstock resistance score and soil susceptibility score on the digital information map are calculated to generate a geospatial risk map.

[0047] The occurrence of diseases is the result of the combined effects of internal factors (the crop's own resistance) and external factors (environmental suitability). By quantifying and integrating these two key static factors, a forward-looking risk assessment can be conducted before the occurrence of diseases, providing a scientific basis for the subsequent allocation of monitoring resources.

[0048] First, using high-precision Geographic Information System (GIS) data, the rootstock variety planted in each grapevine is marked on the map. The built-in plant physiological characteristic database is queried to assign the rootstock a corresponding phylloxera resistance score as the rootstock resistance score. Grape rootstock is the root system of grape plants that has been specially selected, and its core and original function is to control phylloxera.

[0049] By sampling gridded soil, the physicochemical properties of the soil were analyzed to obtain key soil parameters, especially the precise percentage content of clay and sand. Based on soil science knowledge, sandy soil is more conducive to the survival and spread of phylloxera, while plots with high clay content are not conducive to the survival and spread of phylloxera. Therefore, based on the specific parameters of clay and sand in the soil parameters, such as the precise percentage content of clay and sand, standardized soil susceptibility scores were assigned to plots with different soil parameters.

[0050] Using a weighted fusion algorithm, the comprehensive static risk value of each grapevine is calculated as a geospatial risk coefficient, and visualized on a digital information map using color gradients to form a geospatial risk map.

[0051] This invention integrates plant genetic resistance and soil environmental data to construct a quantitative and visualized geospatial risk map. It enables precise prediction and spatial location of potential disease risks, transforming vague agricultural experience into a standardized data model. Compared to traditional extensive management relying on manual inspections, this method scientifically guides the optimal allocation of monitoring resources, prioritizing attention to high-risk areas and significantly improving the probability of early disease detection and management efficiency. It not only provides a crucial static risk benchmark for subsequent dynamic monitoring but also offers long-term decision-making support for planting planning and soil improvement, reducing the likelihood of large-scale disease outbreaks at the source.

[0052] The feature recognition module uses remote sensing technology to periodically collect data on crop plants, obtaining multi-source remote sensing time-series data; it constructs a deep risk recognition model to identify geospatial risk maps, multi-source remote sensing time-series data, and historical status and intervention records to determine plants with abnormal risks.

[0053] During phylloxera infection of grapevines, root damage initially affects water absorption, leading to an abnormally high canopy temperature. Subsequently, it impairs nutrient transport, resulting in decreased chlorophyll content and reduced photosynthetic efficiency. Ultimately, it affects above-ground growth, causing changes in physical structural parameters. Periodic monitoring using a combination of multiple sensors can capture a series of cascading physiological responses triggered by root problems, constructing a comprehensive and dynamic health record.

[0054] Specifically, this involves using a drone equipped with multispectral, hyperspectral, thermal infrared, and lidar sensors to conduct weekly coverage scans of the vineyard along a preset route. After the flight, the system automatically processes the data: using multispectral and hyperspectral data to calculate the normalized difference vegetation index, chlorophyll content, nitrogen state parameters, etc.; using thermal infrared data to generate a canopy temperature map and further analyze water stress status; and using lidar point cloud data to calculate the leaf area index and canopy volume, and using the canopy volume to measure changes in the physical structure of canopy density.

[0055] Spectral data processing: Using multispectral and hyperspectral data, various vegetation indices, such as the Normalized Difference Vegetation Index (NDVI) and chlorophyll index, are calculated using standard formulas, and chlorophyll content and nitrogen status are retrieved based on reflectance in specific bands.

[0056] Thermal infrared data processing: Align thermal infrared images with visible light images, extract the canopy temperature of individual crop plants, compare it with the ambient temperature, and calculate the canopy temperature stress index to assess water stress status.

[0057] LiDAR data processing: Processing point cloud data, through steps such as ground point filtering and single-plant segmentation, to accurately calculate physical structural parameters of each crop plant, such as canopy height, canopy volume, leaf area index, and canopy density.

[0058] Time-series data construction: All the parameters acquired in each acquisition cycle are used as a high-dimensional data vector, i.e., time slices, and arranged in the order of acquisition time to form multi-source remote sensing time-series data for each crop plant.

[0059] This invention integrates a drone platform with multi-source sensors to achieve comprehensive, high-precision, and periodic non-destructive monitoring of crop growth status. It elevates traditional single-point, destructive sampling and detection to macroscopic, non-contact, three-dimensional information perception, greatly expanding data dimensions and acquisition efficiency. The generated multi-dimensional time-series data not only reveals subtle changes in the spectrum, temperature, and physical structure of crop plants, but also captures their dynamic decay characteristics over time, providing a high-quality data foundation for training deep learning models.

[0060] The feature recognition module may further include a multi-disease concurrent recognition unit based on hyperspectral fingerprints;

[0061] The concurrent multi-disease identification unit is used to: utilize hyperspectral sensor-acquired crop canopy hyperspectral reflectance curves covering hundreds of continuous narrow bands; pre-construct a hyperspectral fingerprint feature library containing multiple diseases (such as phylloxera, powdery mildew, and downy mildew) and nutrient stresses (such as nitrogen deficiency and iron deficiency), where each stress type corresponds to one and / or multiple unique spectral absorption / reflectance feature patterns; automatically extract key features from the hyperspectral curves of the crop under test using deep learning algorithms, and perform high-dimensional feature matching with the fingerprint library, thereby enabling the simultaneous identification of multiple concurrent diseases on a single crop and providing confidence scores for each disease; integrate the confidence scores of various diseases into a deep risk identification model to more accurately determine plants with abnormal risks.

[0062] Construction of hyperspectral fingerprint feature library: For crop plants known to be infected with specific diseases (such as phylloxera and powdery mildew) or under specific nutritional stress (such as iron deficiency), a large number of measurements are performed using a hyperspectral instrument, and the features of the measured spectral curves are extracted and labeled to construct a hyperspectral fingerprint feature library.

[0063] Through the above design, the diagnostic breadth and depth of the system have been greatly expanded, upgrading it from a specialized diagnostic system mainly targeting root diseases to one that can simultaneously identify multiple above-ground and underground pests and diseases as well as nutritional problems; it also makes full use of the rich information of hyperspectral data to achieve accurate decoupling and concurrent identification of different stress types.

[0064] This addresses the shortcomings of traditional diagnostic methods, which can only address one problem at a time and are prone to confusing similar symptoms. It provides agricultural managers with a more comprehensive and multi-dimensional understanding of crop health, guiding more complex and integrated field management decisions and significantly improving the precision of agricultural production.

[0065] The identification process of the deep risk identification model includes: using multi-source remote sensing time series data as the core, identifying crop plants whose health indicators show a slow and continuous decline trajectory through time series analysis algorithms, and obtaining dynamic decline characteristics;

[0066] Historical status and intervention records of crop plants are retrieved. Causal correlation analysis is performed based on historical status, intervention time, intervention measures, and intervention effects to obtain historical status correlation coefficients. Geospatial risk maps are used to obtain geospatial risk coefficients of crop plants. A comprehensive analysis is performed based on dynamic decline characteristics, historical status correlation coefficients, and geospatial risk coefficients to obtain a comprehensive risk score. Plants with abnormal risk are identified based on the comprehensive risk score.

[0067] The deep risk identification model is constructed based on a long short-term memory network and is used to identify specific patterns, namely, health indicators exhibiting a slow but continuous decline over several weeks or months, rather than drastic fluctuations caused by short-term drought or mismanagement; it outputs a dynamic decline feature vector, which quantifies the stability and severity of the decline trend.

[0068] Historical state correlation coefficient acquisition: retrieve historical state and intervention records, and calculate the historical state correlation coefficient; this coefficient reflects the degree of abnormality in the crop plant's response to past intervention measures, as well as the similarity between its symptoms and the target disease.

[0069] Geospatial risk coefficient acquisition: The generated geospatial risk map is called, and the geospatial risk coefficient of the crop plant's location is extracted based on the plant's precise geographical coordinates.

[0070] Comprehensive risk score calculation and abnormal crop plant identification: Design a comprehensive analysis model (such as gradient boosting decision tree XGBoost) to fuse the features of the above three dimensions and calculate the comprehensive risk score.

[0071] This invention constructs a deep identification model that integrates dynamic time-series data, historical records, and static spatial risks. Through time-series analysis algorithms, it accurately captures the continuous decline trajectory of crop plant health status. By cross-validating with historical records and geographical risks, it achieves in-depth analysis and confirmation of abnormal signals, significantly improving the accuracy and reliability of risk identification and effectively avoiding misjudgments that may be caused by a single data source.

[0072] The process of obtaining the historical state correlation coefficient includes: obtaining historical state and intervention records to obtain the historical state of crop plants, intervention time, intervention measures and intervention effects; based on the historical state, intervention time and intervention measures in the historical records, querying the built-in agricultural knowledge base to dynamically generate a quantitative standard expected effect; comparing the intervention effect in the historical state and intervention records with the standard expected effect to calculate the effect difference score between the two.

[0073] The historical status records are compared with the database of typical symptoms of phylloxera, and the symptom similarity score is calculated.

[0074] The effect difference score and symptom similarity score are combined and calculated using a preset fusion function to obtain the historical state correlation coefficient. The process for obtaining the historical state correlation coefficient is as follows: Figure 3 As shown.

[0075] This process is based on the principles of causal inference and knowledge graphs. Its core logic is: if a plant’s symptoms are highly similar to those of phylloxera infection and do not produce the expected positive response to routine interventions aimed at addressing these surface symptoms, then these symptoms are likely not caused by common problems (such as lack of fertilizer or water), but by a deeper, more hidden cause (such as root damage caused by phylloxera).

[0076] By quantifying the “differences in treatment effectiveness” and the “similarity of symptoms”, this embodiment provides a logical framework for computer models to assess suspicion; it transforms fragmented historical records into clearly directional, quantifiable evidence, providing crucial historical context information for deep risk identification models.

[0077] Specifically, a structured agricultural knowledge base is pre-built; the agricultural knowledge base includes: a symptom database, which stores descriptions and quantitative characteristics of typical aboveground symptoms of diseases such as phylloxera, such as "yellowing of leaf margins" and "stunted growth of new shoots";

[0078] Intervention-Effect Correlation Library: Stores standard agricultural practices and the standard expected effects that should be produced on healthy crop plants, with the expected effects expressed as quantitative indicators.

[0079] Effect Difference Score Calculation: Historical status and intervention records of the target crop plants are retrieved. Based on the intervention measures recorded, a knowledge base is consulted to obtain the standard expected effect. The actual intervention effect in the records is compared with the standard expected effect to calculate the effect difference score. A higher score indicates that the crop plants are less responsive to conventional treatments and that there are underlying, deeper problems.

[0080] Symptom similarity score calculation: Extract all symptom descriptions from the historical records, such as leaf yellowing and slow growth; compare these symptoms with the typical symptom database of phylloxera in the knowledge base; use a text similarity algorithm of natural language processing (NLP) combined with symptom image or symptom vector similarity to calculate a symptom similarity score, ranging from 0 to 1. The higher the score, the more the historical state matches the typical characteristics of phylloxera.

[0081] Fusion function calculation: By using a preset fusion function, the effect difference score and symptom similarity score are combined to calculate the final historical state correlation coefficient.

[0082] This invention achieves in-depth mining and intelligent analysis of crop plant historical health records through dynamic querying of a built-in agricultural knowledge base; it transforms qualitative historical descriptions into quantitative, diagnostically valuable correlation coefficients, solving the problem that historical data is difficult for models to directly utilize; by comparing actual intervention effects with standard expected effects, it can accurately identify crop plants that respond abnormally to routine management measures, revealing the underlying causes that may exist beneath the surface symptoms; and combined with similarity analysis with typical disease symptoms, it provides strong historical evidence to support early warning of root diseases, significantly enhancing the logical reasoning ability and accuracy of the diagnostic model.

[0083] The calculation process for the effect difference score can be further defined as follows:

[0084] The system incorporates a machine learning-based dynamic expected effect generation unit to replace the static standard expected effects in the built-in agricultural knowledge base. When historical intervention records of crop plants are obtained, this unit not only retrieves the type of intervention but also simultaneously acquires the physiological state represented by multidimensional sensor data of the crop plant at the intervention time point, as well as environmental meteorological data for a period of time after the intervention. Based on the above multidimensional inputs, the unit dynamically predicts and generates a personalized expected health trajectory curve that the crop plant should have under specific conditions. The system then uses the subsequently periodically collected actual health status data to form the actual health trajectory curve and compares it with the personalized expected trajectory curve. The system calculates the morphological differences and separation degree of the two curves using a time series similarity algorithm, thereby generating the final effect difference score.

[0085] Training the dynamic prediction model: A dataset containing "pre-intervention state - intervention measures - post-intervention environment - actual intervention effect trajectory" is pre-collected. This data is used to train a time series prediction model; the model learns the dynamic effects that various intervention measures should produce under different initial conditions and environments.

[0086] Personalized Expected Health Trajectory Generation: The dynamic expected effect generation unit is triggered when the system analyzes historical records; (The following information is retrieved:)

[0087] Intervention measures: Apply nitrogen fertilizer;

[0088] Initial state: Multidimensional data of crop plant A on June 1 (e.g., NDVI = 0.65, canopy temperature = 28℃, leaf area index = 2.1);

[0089] Environmental data: Weather forecast data from June 1 to June 15 (such as sunshine, temperature, and rainfall).

[0090] Based on the above inputs, the model predicts a vegetation index change curve for the next 14 days; for example, the vegetation index is expected to start to rise steadily after 3 days, reach a peak of 0.72 on the 10th day, and then slowly stabilize.

[0091] Actual trajectory acquisition and comparison: The system uses subsequent remote sensing monitoring to obtain the actual vegetation index change curve of crop plant A from June 1st to June 15th. For example, the actual vegetation index rose slightly to 0.67 in the first 5 days, but then began to decline.

[0092] Difference score calculation: A dynamic time warping algorithm is used to calculate the warped path distance between the personalized expected health trajectory curve and the actual health trajectory curve. The dynamic time warping algorithm can effectively measure the similarity between two time series of unequal durations, even with some delay or scaling on the time axis. The calculated distance is normalized to obtain the final effect difference score; a larger score means that the actual recovery deviates significantly from the ideal recovery under the current conditions.

[0093] This design creatively upgrades the evaluation of effect differences from a static, single-point numerical comparison to a dynamic, multi-point, full-process trajectory morphology comparison. Its core advantage lies in:

[0094] Highly contextualized and personalized: By dynamically generating expected results, key factors such as the crop's own health foundation and environmental variables during the intervention period are taken into account, and a highly scientific and personalized benchmark is tailored for each intervention.

[0095] Enhanced diagnostic information: By comparing the trajectory of the entire recovery process, the system can capture richer and more subtle abnormal signals. For example, it can distinguish between different patterns of abnormal responses such as "completely ineffective," "rising first and then falling," or "extremely slow recovery," which may correspond to different underlying causes, thus providing deeper insights for diagnosis.

[0096] The robustness and accuracy are significantly enhanced: This method can effectively filter out crop plants that do not respond well due to objective reasons such as bad weather or poor crop plant condition, providing extremely reliable evidence for identifying deep and chronic diseases.

[0097] The infection assessment module collects data from at-risk and abnormal plants using acoustic detection equipment to obtain acoustic feature data; identifies at-risk and abnormal plants using a ring electrode array to generate a three-dimensional spatial distribution map of the root system's electrical properties; and identifies the abnormal state index of the crop plant based on the acoustic feature data and the three-dimensional spatial distribution map of the electrical properties.

[0098] The process of generating a three-dimensional electrical property spatial distribution map of the root system includes: deploying a ring electrode array in the soil around the base of the plant at risk and abnormality; applying weak, multi-frequency high-frequency alternating current sequentially between different electrode pairs through the host system; and simultaneously measuring the voltage distribution on all other electrodes to collect and form a complete electrical response dataset.

[0099] The electrical response dataset is solved by inverse operation using a reconstruction algorithm to generate a three-dimensional matrix in units of voxels. Based on the three-dimensional matrix, a three-dimensional spatial distribution map of electrical characteristics is obtained, including the three-dimensional coordinates, impedance value, and capacitance characteristic parameters obtained by multi-frequency scanning for each voxel.

[0100] This process is based on bioelectrical impedance tomography (EIT) technology. Its core physical principle is that different biological tissues possess different electrical properties, such as resistivity and dielectric constant. Healthy plant root tissue, tissue infected with diseases (such as phylloxera) and decaying, as well as the surrounding soil medium, all exhibit significantly different cellular structures, water contents, and ion concentrations, thus displaying markedly different electrical impedance values. By applying an excitation current around the root system and measuring the boundary voltage, the EIT system can acquire information reflecting the distribution of internal conductivity.

[0101] The reconstruction algorithm is a mathematical "inverse problem" solution process that can inversely determine the internal attribute distribution from external measurements. Multi-frequency scanning is chosen to obtain richer tissue information because different frequencies of current have different sensitivities to cell membranes (capacitive properties) and cellular fluid (resistive properties), which helps to more accurately distinguish between healthy and diseased tissues.

[0102] The "adjacent excitation-adjacent measurement" mode is adopted, in which a pair of adjacent electrodes are selected as excitation electrodes, and a weak, multi-frequency high-frequency sinusoidal alternating current is applied. While the excitation is applied, the voltage distribution on all adjacent electrode pairs is measured simultaneously.

[0103] The excitation electrode pairs are automatically switched in sequence, and the above process is repeated until all possible excitation measurement combinations are completed. The entire process forms a complete dataset containing multi-frequency and multi-angle electrical responses.

[0104] Image Reconstruction: The acquired electrical response dataset is input into a computer equipped with an image reconstruction algorithm. An inverse problem-solving algorithm based on the finite element method, such as the Gauss-Newton algorithm, is employed. First, a three-dimensional finite element mesh model matching the actual detection area is established. The reconstruction algorithm iteratively optimizes the calculation of the impedance value of each element (voxel) within the mesh, ensuring that the boundary voltage calculated through the forward problem under this impedance distribution best matches the actually measured voltage data.

[0105] Finally, the algorithm outputs a three-dimensional matrix in units of voxels. In this matrix, each voxel contains its three-dimensional spatial coordinates, the impedance value at that location, and capacitance characteristic parameters (such as phase angle) obtained from multi-frequency scanning.

[0106] 3D Image Visualization: Using 3D visualization software, the reconstructed 3D matrix is ​​rendered into an intuitive 3D spatial distribution map of electrical properties. In the image, different impedance values ​​can be represented by different colors and transparency, thus clearly showing the spatial distribution of the electrical properties of the root system and the surrounding soil medium.

[0107] This invention applies electrical impedance tomography (EIT) technology to non-destructive testing of plant roots, enabling precise three-dimensional and visual diagnosis of the health status of underground roots. The generated three-dimensional spatial distribution map of electrical characteristics can intuitively display the location, morphology, and approximate extent of diseased areas, with an accuracy down to the centimeter level. Capacitance parameters obtained through multi-frequency scanning further enhance the ability to distinguish different tissue types. It overcomes the limitations of traditional root detection methods, which rely on excavation and sampling, are highly destructive, and cannot be repeated, providing a dynamic and effective observation method.

[0108] The process of identifying the abnormal state index of crop plants includes: monitoring the elastic waves generated in the xylem vessels of crop plants under water stress due to the rupture of the water column, and obtaining acoustic characteristic data;

[0109] Acoustic correlation feature data is obtained by matching the acoustic feature data with the preset acoustic spectrum library of phylloxera threat. Based on the acoustic correlation feature data and the three-dimensional coordinates, impedance values ​​and capacitance parameters of voxels in the three-dimensional electrical characteristic spatial distribution map, the crop plants infected by phylloxera are identified. The acoustic spectrum library of phylloxera threat is obtained by acoustic detection of phylloxera-infected plants.

[0110] The spatial distribution map of three-dimensional electrical properties is processed by a three-dimensional image segmentation algorithm. Based on the impedance value and the three-dimensional morphology of the diseased tissue, the total volume of the diseased tissue and the total volume of the healthy root system are identified and calculated, and the abnormal state index is calculated.

[0111] This process is based on the principle of multi-physics coupling diagnosis, combining two independent physical detection methods to cross-verify and improve the accuracy of diagnosis.

[0112] Acoustic detection: Based on plant physiology, root damage leads to impaired water absorption, which can cause xylem plugging under water stress, producing unique acoustic emission signals; this is a functional detection method to determine whether the root system is malfunctioning.

[0113] EIT detection: Based on biophysics, it directly performs three-dimensional imaging of the root system to observe its structure; it is a morphological detection method to determine where there are abnormalities in the root system.

[0114] Combining functional abnormality evidence from acoustic testing with structural abnormality evidence from EIT creates a complete diagnostic loop, significantly reducing the interference of other stress factors. Finally, through 3D image segmentation and volume calculation, the qualitative diagnostic results are transformed into a precise, standardized quantitative indicator—the Abnormal State Index—which is crucial for assessing the condition, guiding treatment, and predicting yield loss.

[0115] Specifically, acoustic feature data acquisition: A highly sensitive piezoelectric acoustic emission sensor is installed at the base of the main stem of the plant with abnormal risk. It is responsible for monitoring and recording the weak elastic wave (stress wave) signal generated in the xylem vessels of the crop plant due to the breakage of the water column. The acquisition frequency is 20-200kHz to form the original acoustic feature data.

[0116] Acoustic data analysis and matching: The acquired acoustic signals are filtered and feature extracted, and parameters such as amplitude, duration, and energy are calculated. The extracted signal features are then matched with a pre-defined acoustic spectrum library of phylloxera threats. This library pre-stores a large number of cavitation acoustic signal patterns unique to root water absorption dysfunction caused by phylloxera infestation, as confirmed in experiments.

[0117] Acoustic correlation feature data is obtained by calculating the matching degree between the measured signal and feature signals in the database using pattern recognition algorithms such as Dynamic Time Warping (DTW). A high matching degree indicates that the detected water stress is highly likely caused by phylloxera.

[0118] Multimodal data comprehensive analysis and infection judgment: Acoustic analysis results serve as key verification information. If the crop plant simultaneously exhibits a high impedance abnormal area (from EIT) and a high-match stress acoustic signal, the system judges with extremely high confidence that the crop plant is infected by phylloxera.

[0119] Abnormal state index calculation: The spatial distribution map of the three-dimensional electrical characteristics of the diagnosed crop plants is processed. First, a three-dimensional image segmentation algorithm (such as threshold-based segmentation or more advanced deep learning segmentation models such as U-Net) is used to accurately divide the voxels in the three-dimensional image into three categories: healthy roots, diseased tissue, and soil background, based on the difference in electrical impedance values ​​between healthy roots and diseased tissue (diseased tissue usually has higher electrical impedance).

[0120] Based on the segmentation results, the total volume of all voxels identified as diseased tissues and the total volume of all voxels identified as healthy roots are automatically calculated, and the abnormal state index is finally quantified.

[0121] This invention establishes a high-confidence method for diagnosing root diseases through the synergistic analysis of acoustic detection and three-dimensional electrical property spatial distribution maps. Matching acoustic signals provides physiological functional evidence for abnormal areas discovered in the three-dimensional electrical property spatial distribution maps, significantly improving the specificity and accuracy of diagnosis and effectively avoiding misdiagnosis. Furthermore, through intelligent segmentation and calculation of three-dimensional images, precise and non-destructive quantification of the degree of root infection is achieved. This provides a scientific basis for assessing the severity of disease in individual crop plants and a unified data foundation for subsequent regional disease assessment and control decisions.

[0122] The anomaly warning module uses cluster analysis based on the abnormal state index and three-dimensional coordinates of crop plants to determine the infection center and infection area, and then issues anomaly warnings.

[0123] The process of identifying infection centers and infection areas includes: binding the abnormal state indices of all confirmed crop plants with three-dimensional geographic coordinates to form a spatial point dataset; using a density-based spatial clustering algorithm to analyze the spatial point dataset and identify crop plants that are spatially adjacent and whose infection indices all reach a preset high-risk threshold as a cluster;

[0124] The geometric center of the cluster with the highest density is determined as the infection center, and the convex hull and contour boundary of the geographical area covered by the cluster at the infection center are used to determine the infection area.

[0125] This process is based on the principles of geographic information science and spatial statistics, especially the theory of spatial autocorrelation, which states that geographically proximate things are more likely to have similar attributes. Disease spread is usually not random, but rather clustered outwards from one or more source points. The core advantage of density-based spatial clustering algorithms lies in their ability to discover density-based clustering patterns. Through this algorithm, scattered single-plant diagnostic information can be elevated to a regional, spatially patterned understanding of disease trends.

[0126] Identifying the infection center is to find the core point where the disease is most severe and most likely to be the source of spread, so as to facilitate source tracing analysis and centralized treatment; identifying the infection area is to define a clear management boundary, guide the scope of agricultural operations (such as isolation and precision application), and prevent the further spread of the disease.

[0127] The anomaly warning module also includes a spatiotemporal diffusion prediction unit;

[0128] The spatiotemporal diffusion prediction unit is used to: comprehensively call upon the identified infection centers and areas, geospatial risk maps, real-time meteorological data and soil parameter distribution maps obtained from external sources; based on a preset disease diffusion model, simulate and predict the most likely diffusion direction, speed and range of the infected area in the future; and finally, on the early warning map, in addition to marking the current infected area, visualize the potential high-risk spread areas in the future in the form of a dynamic and gradual risk layer.

[0129] Multi-source data integration: After the spatiotemporal diffusion prediction unit is activated, it first acquires the currently identified infection centers, then loads the GSR geospatial risk map as the "resistance / gravity" background for diffusion. Finally, it acquires wind speed / direction data from local weather stations and humidity data from the soil moisture monitoring network in real time via API interface.

[0130] Diffusion model simulation: The entire plot is gridded using a cellular automata model, and the state (healthy / latent / infected) of each grid cell is iteratively updated based on its own state, the states of its neighbors, and a series of transition rules.

[0131] Rule definition: The probability of diffusion in the downwind direction is much higher than in the upwind direction; in areas with high GSR values ​​(such as sandy soil), the diffusion speed is faster; in areas where the soil is too dry or too wet, the diffusion speed will slow down.

[0132] Prediction and Visualization: The model iterates forward for 14 cycles (representing 14 days) to generate daily spatial distribution prediction maps of the disease. On the final warning map, these 14-day predictions are overlaid, with red areas of varying transparency representing the potential infection range at different future time points, forming a dynamic risk corridor.

[0133] This design significantly enhances the foresight and proactiveness of the early warning system, upgrading traditional static alarms to dynamic prediction. By integrating multi-physics data and simulating disease propagation dynamics, it predicts where diseases will go. It provides proactive intervention strategies to effectively block disease transmission chains, achieving an intelligent upgrade from risk management to future risk prediction.

[0134] This invention combines diagnostic data from individual crop plants with geospatial information and utilizes a density-based clustering algorithm to achieve a leap from point-based diagnosis to area-based state potential perception. It can automatically and objectively identify the spatial distribution pattern of diseases in the field, accurately locate infection centers, and delineate the boundaries of infection areas. By implementing a zoned management strategy, it achieves a shift from passive and comprehensive prevention and control to proactive and precise regional control, significantly improving prevention and control efficiency and reducing resource waste.

[0135] Example 2:

[0136] This invention proposes an intelligent diagnostic system for plants and animals based on multi-source data. Building upon Embodiment 1, specific implementation details are provided, including:

[0137] The data acquisition module identifies geospatial risk maps based on rootstock and soil parameters; it also acquires historical status and intervention records of crop plants, obtaining information on the historical status of crop plants, intervention time, intervention measures, and intervention effects.

[0138] The feature recognition module uses remote sensing technology to periodically collect data on crop plants and obtain multi-source remote sensing time-series data; it constructs a deep risk recognition model to identify geospatial risk maps, multi-source remote sensing time-series data, and historical status and intervention records to determine plants with abnormal risks.

[0139] The infection assessment module collects data from at-risk and abnormal plants using acoustic detection equipment to obtain acoustic feature data; identifies at-risk and abnormal plants using a ring electrode array to generate a three-dimensional spatial distribution map of the root system's electrical properties; and identifies the abnormal state index of the crop plant based on the acoustic feature data and the three-dimensional spatial distribution map of the electrical properties.

[0140] The anomaly warning module uses cluster analysis based on the abnormal state index and three-dimensional coordinates of crop plants to determine the infection center and infection area, and then issues anomaly warnings.

[0141] Specifically, in the GIS platform, the digital orthophoto map, the vector planting point layer containing the rootstock type, and the vector plot layer of soil type (e.g., sandy loam, clay loam) are spatially registered and overlaid.

[0142] Values ​​are assigned to different rootstocks based on a built-in database of plant physiological characteristics. Values ​​are also assigned to different soil types based on soil physicochemical parameters. For example, sandy loam (high risk) is assigned a value of 8.5, and clay loam (low risk) is assigned a value of 3.0.

[0143] For each crop plant point on the map, its geospatial risk coefficient is calculated using a weighted fusion algorithm.

[0144] Using the Kriging spatial interpolation method, a smooth risk level raster layer covering the entire monitoring area is generated based on the geospatial risk coefficients of all points; the layer visually displays the static risk distribution from low to high using a color spectrum (green-yellow-red).

[0145] Using a drone platform, drones equipped with multispectral sensors, hyperspectral sensors, thermal infrared sensors, and lidar sensors are used to periodically scan crop plants, collecting data in 7-day cycles during the critical growth periods of the crop plants.

[0146] Vegetation index, chlorophyll content, and nitrogen status parameters were obtained using multispectral and hyperspectral sensors; canopy temperature was obtained using thermal infrared sensors to assess water stress caused by root damage; and canopy three-dimensional point cloud data was obtained using lidar sensors to calculate leaf area index and canopy density physical structure parameters.

[0147] The multi-dimensional data obtained in each acquisition cycle are used as time slices, and multi-source remote sensing time series data are formed according to the acquisition order.

[0148] When the aforementioned time series data is input into a pre-trained deep risk identification model, the model will output a high-score dynamic decline feature for crop plants exhibiting a continuous and slow decline.

[0149] The deep risk identification model is built around a Long Short-Term Memory (LSTM) network, containing two hidden layers, each with 64 units. The input data consists of feature vectors from multi-source remote sensing data collected over multiple consecutive periods. The LSTM output is a feature value quantifying the decline trend; this feature value, along with the historical state correlation coefficient and the geospatial risk coefficient, forms a 3D feature vector, which is input to a pre-trained XGBoost classifier. This classifier contains 150 estimators with a maximum depth of 4. The model outputs a comprehensive risk score between 0 and 1; when the score is greater than 0.8, the plant is identified as a plant with abnormal risk.

[0150] Historical causal association analysis: The system automatically retrieves historical intervention records, compares these records with those in the agricultural knowledge base, and calculates the difference in effectiveness scores; combined with symptom similarity analysis, it ultimately generates historical state correlation coefficients. A fusion model integrates information from various dimensions to calculate a comprehensive risk score.

[0151] Three-dimensional electrical impedance tomography (EIT): A ring electrode array was deployed on high-risk crop plants, and multi-frequency scanning from 1 kHz to 100 kHz was performed; a reconstruction algorithm generated a three-dimensional voxel model of the electrical impedance distribution in the root region. The model showed that the electrical impedance value of healthy root tissue was approximately 500 Ω·m, while large areas of high impedance anomalies exceeding 1500 Ω·m appeared in the model.

[0152] Cross-validation of acoustic emission (AE) signals: AE sensors were deployed on the main stem of crop plants, and several acoustic signals of xylem plugging induced by water stress were detected. The peak frequency (approximately 150 kHz) and energy distribution of the signals matched the acoustic spectrum library of phylloxera stress by 92%, providing functional evidence for the structural anomalies detected by EIT.

[0153] Abnormal state index calculation: The 3DU-Net segmentation algorithm is used to intelligently segment the EIT three-dimensional model and automatically calculate the total volume of diseased tissue and the total volume of healthy roots to calculate the final quantitative index.

[0154] Spatial point dataset generation: The geographic coordinates of all confirmed crop plants are bound to anomaly indices to form a spatial point dataset; density-based clustering analysis is used to calculate the geometric centroid of the cluster, which serves as the core point of the disease outbreak. Infected area: The minimum convex hull surrounding all points in the cluster is generated and defined as the core infected area. Based on this, a final control and management area containing a safety buffer zone is generated.

[0155] On the GIS early warning platform, the system highlights the infection center with a high-brightness icon, displays the prevention and control management area as a semi-transparent polygon layer, and provides clear prevention and control suggestions, thus completing the entire intelligent diagnosis and early warning closed loop.

[0156] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent diagnostic system for plants and animals based on multi-source data, characterized in that, include: The data acquisition module obtains a geospatial risk map based on rootstock parameters and soil parameters; Obtain historical status and intervention records of crop plants to obtain information on the historical status of crop plants, intervention time, intervention measures, and intervention effects; The identification process of the geospatial risk map includes: Obtain digital information maps and mark the rootstock parameters of crop plants and the soil parameters of the plots they are located on; Based on the plant physiological characteristics database, a standardized rootstock resistance score was assigned to each rootstock, and a standardized soil susceptibility score was assigned to plots with different soil parameters based on the percentage content of clay and sand in the soil parameters. By using a weighted fusion algorithm, the rootstock resistance score and soil susceptibility score on the digital information map are calculated to generate a geospatial risk map. The feature recognition module uses remote sensing technology to periodically collect data on crop plants and obtain multi-source remote sensing time-series data; it constructs a deep risk recognition model to identify geospatial risk maps, multi-source remote sensing time-series data, and historical status and intervention records to determine plants with abnormal risks. The identification process of the deep risk identification model includes: using multi-source remote sensing time series data as the core, identifying crop plants with continuously declining health indicators through time series analysis algorithms, and obtaining dynamic decline characteristics. Historical status and intervention records of crop plants are retrieved. Causal correlation analysis is performed based on historical status, intervention time, intervention measures, and intervention effects to obtain historical status correlation coefficients. Geospatial risk maps are used to obtain geospatial risk coefficients of crop plants. A comprehensive analysis is performed based on dynamic decline characteristics, historical status correlation coefficients, and geospatial risk coefficients to obtain a comprehensive risk score. Plants with abnormal risk are identified based on the comprehensive risk score. The infection assessment module collects data from at-risk and abnormal plants using acoustic detection equipment to obtain acoustic feature data; identifies at-risk and abnormal plants using a ring electrode array to generate a three-dimensional spatial distribution map of the root system's electrical properties; and identifies the abnormal state index of the crop plant based on the acoustic feature data and the three-dimensional spatial distribution map of the electrical properties. The process of identifying the abnormal state index of crop plants includes: monitoring the elastic waves generated in the xylem vessels of crop plants under water stress due to the rupture of the water column, and obtaining acoustic characteristic data; The acoustic feature data is matched with the pre-set acoustic spectrum library of phylloxera threat to obtain acoustic correlation feature data. Based on the acoustic correlation feature data and the three-dimensional coordinates, impedance values ​​and capacitance parameters of voxels in the three-dimensional electrical characteristic spatial distribution map, the crop plants infected by phylloxera are identified. The spatial distribution map of three-dimensional electrical properties is processed by a three-dimensional image segmentation algorithm. Based on the impedance value and the three-dimensional morphology of the diseased tissue, the total volume of the diseased tissue and the total volume of the healthy root system are identified and calculated, and the abnormal state index is calculated. Acoustic feature data acquisition: A highly sensitive piezoelectric acoustic emission sensor is installed at the base of the main stem of the plant with abnormal risk. It is responsible for monitoring and recording the stress wave signal generated in the xylem vessels of the crop plant due to the breakage of the water column, forming the original acoustic feature data. Acoustic data analysis and matching: The collected acoustic signals are filtered and feature extracted, and their amplitude, duration and energy parameters are calculated; the extracted signal features are matched with the preset acoustic spectrum library of phylloxera threat; the acoustic spectrum library of phylloxera threat pre-stores the cavitation acoustic signal patterns that are unique to root water absorption dysfunction caused by phylloxera infestation. The anomaly warning module uses cluster analysis based on the abnormal state index and three-dimensional coordinates of crop plants to determine the infection center and infection area, and then issues anomaly warnings.

2. The intelligent plant and animal diagnostic system based on multi-source data according to claim 1, characterized in that: The acquisition methods for the multi-source remote sensing time series data include: using a drone equipped with a multispectral sensor, a hyperspectral sensor, a thermal infrared sensor, and a lidar sensor to perform periodic remote sensing scans of crop plants; Among them, vegetation index, chlorophyll content, and nitrogen status parameters are obtained through multispectral and hyperspectral sensors; canopy temperature is obtained through thermal infrared sensors to assess water stress caused by root damage; and canopy three-dimensional point cloud data is obtained through lidar sensors to calculate leaf area index and canopy density physical structure parameters. The multi-dimensional data obtained in each acquisition cycle are used as time slices, and multi-source remote sensing time series data are formed according to the acquisition order.

3. The intelligent plant and animal diagnostic system based on multi-source data according to claim 1, characterized in that: The process of obtaining the historical state correlation coefficient includes: obtaining historical state and intervention records to obtain the historical state of crop plants, intervention time, intervention measures and intervention effects; based on the historical state, intervention time and intervention measures, querying the built-in agricultural knowledge base to dynamically generate a quantitative standard expected effect; comparing the intervention effect in the historical state and intervention records with the standard expected effect to calculate the effect difference score. The historical status records are compared with the database of typical symptoms of phylloxera, and the symptom similarity score is calculated. By using a preset fusion function, the effect difference score and symptom similarity score are combined and calculated to obtain the historical state correlation coefficient.

4. The intelligent plant and animal diagnostic system based on multi-source data according to claim 1, characterized in that: The process of generating a three-dimensional electrical property spatial distribution map of the root system includes: deploying a ring electrode array in the soil at the base of the plant at risk and abnormality, applying weak, multi-frequency high-frequency alternating current sequentially between different electrode pairs through the host system, and simultaneously measuring the voltage distribution on all other electrodes to collect and form a complete electrical response dataset. The electrical response dataset is solved by inverse operation using a reconstruction algorithm to generate a three-dimensional matrix in units of voxels. Based on the three-dimensional matrix, a three-dimensional spatial distribution map of electrical characteristics is obtained, including the three-dimensional coordinates, impedance value, and capacitance characteristic parameters obtained by multi-frequency scanning for each voxel.

5. The intelligent plant and animal diagnostic system based on multi-source data according to claim 1, characterized in that: The process of identifying infection centers and infection areas includes: binding the abnormal state index of crop plants with three-dimensional geographic coordinates to form a spatial point dataset; using a density-based spatial clustering algorithm to analyze the spatial point dataset and identify crop plants that are spatially adjacent and whose infection indices all reach a preset high-risk threshold as a cluster; The geometric center of the cluster with the highest density is determined as the infection center, and the convex hull and contour boundary of the geographical area covered by the cluster at the infection center are used to determine the infection area.

Images