Accurate targeted early warning method and device for meteorological disaster of kiwifruit canker and medium

Abstract

The invention relates to the technical field of data analysis, and discloses a precise targeted early warning method and device for meteorological disasters of kiwifruit canker and a medium. The method comprises the following steps: performing standardization processing on historical occurrence time, morbidity degree and distribution range data of kiwifruit canker as well as meteorological observation data and cultivation management data of corresponding time periods to obtain a historical disease database; performing multivariate statistical analysis processing on the three-dimensional meteorological monitoring network data and the historical disease database to obtain a disease risk assessment index; performing feature extraction processing on the disease risk assessment indexes through a deep learning algorithm to obtain disease early warning parameters; performing gridding partition processing on the disease early warning parameters to obtain partition early warning data; and performing dynamic optimization processing on the accuracy, the advance and the coverage rate of the partition early warning data to obtain accurate targeted early warning information of the meteorological disaster of the kiwifruit canker. According to the application, the accuracy and timeliness of the existing kiwi fruit canker meteorological disaster early warning method are improved.

Description

Technical Field

[0001] The present application relates to the field of data analysis, and in particular to a precise targeted early warning method, device and medium for meteorological disasters of kiwifruit canker. Background Art

[0002] Kiwifruit canker is one of the most destructive diseases in kiwifruit production, seriously affecting the healthy development of the kiwifruit industry. At present, the early warning of kiwifruit canker at home and abroad mainly relies on traditional meteorological monitoring and manual experience judgment. These methods include setting up fixed meteorological observation points to collect data such as temperature and humidity, combining historical disease patterns for prediction, or conducting prevention and control after discovering early symptoms of the disease through field inspections. At the same time, some regions have begun to try to use computer technology and mathematical models to predict the occurrence of diseases, but the early warning system is still in its infancy.

[0003] However, the existing early warning methods have many shortcomings: first, traditional fixed observation points are difficult to accurately reflect microclimate changes under complex terrain conditions, resulting in low warning accuracy; second, methods that rely solely on manual experience and judgment are highly subjective and difficult to detect potential risks in a timely manner; third, existing early warning models often ignore the impact of topography on local meteorological conditions, and fail to make full use of multi-source data for comprehensive analysis, resulting in insufficient temporal and spatial accuracy of warnings; finally, the way of releasing warning information is single, making it difficult to achieve accurate push and rapid response. Summary of the invention

[0004] The present application provides a precise targeted early warning method, device and medium for meteorological disasters of kiwifruit canker, which is used to achieve precise early warning of kiwifruit canker by building a complete monitoring-assessment-early warning-feedback system. It has strong practicality and innovation, fully considers the geographical and climatic characteristics and the development needs of the kiwifruit industry, and can provide fruit farmers with timely and accurate disease prevention and control guidance.

[0005] In the first aspect, the present application provides a precise targeted early warning method for meteorological disasters caused by kiwifruit canker, which comprises: performing real-time Internet of Things transmission processing on the temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by layered meteorological monitoring stations to obtain three-dimensional meteorological monitoring network data; standardizing the historical occurrence time, severity of disease, distribution range data of kiwifruit canker, and meteorological observation data and cultivation management data of the corresponding time period to obtain a historical disease database; performing multivariate statistical analysis processing on the three-dimensional meteorological monitoring network data and the historical disease database to obtain disease risk assessment indicators; performing feature extraction processing on the disease risk assessment indicators through a deep learning algorithm to obtain disease warning parameters; performing grid partitioning processing on the disease warning parameters to obtain partition warning data; dynamically optimizing the accuracy, lead time and coverage of the partition warning data to obtain precise targeted early warning information on meteorological disasters caused by kiwifruit canker.

[0006] In the second aspect, the present application provides a precise targeted early warning device for meteorological disasters caused by kiwifruit canker, and the precise targeted early warning device for meteorological disasters caused by kiwifruit canker comprises:

[0007] The transmission module is used to perform real-time Internet of Things transmission processing on the temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring station to obtain three-dimensional meteorological monitoring network data;

[0008] The processing module is used to standardize the historical occurrence time, disease severity, distribution range data of kiwifruit canker disease, as well as the meteorological observation data and cultivation management data of the corresponding period to obtain a historical disease database;

[0009] An analysis module, used to perform multivariate statistical analysis on the three-dimensional meteorological monitoring network data and the historical disease database to obtain disease risk assessment indicators;

[0010] An extraction module is used to perform feature extraction processing on the disease risk assessment index through a deep learning algorithm to obtain disease warning parameters;

[0011] A partitioning module is used to perform grid partitioning processing on the disease warning parameters to obtain partition warning data;

[0012] The optimization module is used to dynamically optimize the accuracy, lead time and coverage of the sub-area warning data to obtain precise and targeted warning information on kiwifruit canker meteorological disasters.

[0013] The third aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned precise targeted early warning method for meteorological disasters caused by kiwifruit canker.

[0014] In the technical solution provided by the present application, the temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring stations are processed in real time through the Internet of Things, so as to realize the three-dimensional collection and real-time monitoring of multi-level meteorological data, establish a complete three-dimensional meteorological monitoring network data, provide timely and accurate basic data support for disease early warning, and standardize the historical occurrence time, severity and distribution range data of kiwifruit canker disease and the meteorological observation data and cultivation management data of the corresponding time period to establish a systematic historical disease database, which provides reliable historical data for disease law analysis. With data support, multivariate statistical analysis and processing are performed on the three-dimensional meteorological monitoring network data and historical disease database to extract disease risk assessment indicators, accurately identify the combination of key factors that cause the disease, and use deep learning algorithms to perform feature extraction on disease risk assessment indicators to obtain disease warning parameters with predictive capabilities, thereby improving the accuracy and scientificity of the warning. Grid-based zoning of disease warning parameters is then used to achieve spatial refinement of warning information. Finally, the accuracy, lead time and coverage of the zoned warning data are dynamically optimized to continuously improve the warning effect, ultimately achieving accurate warning of kiwifruit canker. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 This is a schematic diagram of an embodiment of a method for accurately targeting early warning of meteorological disasters caused by kiwifruit canker in an embodiment of the present application;

[0017] Figure 2 This is a schematic diagram of an embodiment of a precise targeted early warning device for kiwifruit canker meteorological disasters in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The embodiments of the present application provide a method, device and medium for accurately targeting early warning of meteorological disasters of kiwifruit canker. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the precise targeted early warning method for meteorological disasters of kiwifruit canker disease includes:

[0020] Step S101, the temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring station are processed by the Internet of Things in real time transmission to obtain three-dimensional meteorological monitoring network data;

[0021] Step S102, standardizing the historical occurrence time, disease severity, and distribution range data of kiwifruit canker disease, as well as the meteorological observation data and cultivation management data of the corresponding time period, to obtain a historical disease database;

[0022] Step S103, performing multivariate statistical analysis on the three-dimensional meteorological monitoring network data and the historical disease database to obtain disease risk assessment indicators;

[0023] Step S104: extracting features from the disease risk assessment indicators using a deep learning algorithm to obtain disease warning parameters;

[0024] Step S105, gridding and partitioning the disease warning parameters to obtain partition warning data;

[0025] Step S106: dynamically optimize the accuracy, lead time and coverage of the zone warning data to obtain precise and targeted warning information on the meteorological disaster of kiwifruit canker.

[0026] It is understandable that the execution subject of the present application can be a precise targeted early warning device for meteorological disasters of kiwifruit canker, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0027] Specifically, the Internet of Things is used for real-time transmission processing. Basic meteorological data are collected through layered meteorological monitoring stations installed at different altitudes. The version number and operating status of the collected meteorological monitoring station equipment are detected, and the equipment status data is generated and monitored in real time to form equipment operating parameters. The abnormal equipment is screened out by setting threshold analysis, and the valid equipment list is obtained after the abnormal data is eliminated. According to the valid equipment list, temperature, humidity, precipitation, wind direction, wind speed, and light data are collected to form the first meteorological data set, and soil moisture data and canopy temperature and humidity data are collected to form the second meteorological data set. The quality of these two data sets is evaluated, and the data with a score lower than 85 points is eliminated to obtain the cleaned meteorological data. Subsequently, the time series interpolation algorithm is used to fill in the missing data values, and the completed data is standardized and converted. Finally, the multi-source data is integrated into a three-dimensional meteorological monitoring network data through the data fusion algorithm. In the historical data processing link, the format of the historical occurrence time data of kiwifruit canker is unified, and the time data of different recording methods are converted into a standard format. The time period characteristics are obtained through seasonal analysis. The disease severity data was quantified and divided into three levels: mild, moderate, and severe, with values ​​of 1, 2, and 3 respectively. The disease distribution characteristics were obtained by combining the spatial distribution information. The disease distribution characteristics were correlated with the time period characteristics, the correlation coefficient was calculated, the spatiotemporal correlation index was determined, and weight coefficients were assigned to different indicators to obtain the disease characteristic weights. The meteorological observation data was normalized, and the data of different dimensions were unified to the interval of 0-1. The missing values ​​were filled to obtain complete meteorological data. The cultivation management data such as fertilization and pruning were encoded and converted to quantify the impact of management measures, and the correlation was analyzed to obtain the management impact factor. Finally, the complete meteorological data, disease characteristic weights, and management impact factors were integrated into a historical disease database through a data fusion algorithm.

[0028] In the multivariate statistical analysis stage, the temperature data in the three-dimensional meteorological monitoring network data is periodically analyzed to obtain the temperature change law, set the temperature threshold range, and calculate the temperature risk index based on the high risk of disease when the temperature is in the range of 15-25℃. The duration of humidity data is analyzed, the duration of relative humidity greater than 85% is counted, the humidity risk level is assessed, and the humidity risk index is generated. The temperature risk index and the humidity risk index are coupled and analyzed to construct an environmental combination index, verify the accuracy of the threshold, and determine the key environmental factors. The disease law is extracted from the historical disease database, the frequency of disease in different periods is calculated, the characteristics of the high incidence period are obtained, the characteristics of the high incidence period are matched with the key environmental factors, and the risk prediction factors are determined. The weights of each factor are calculated, the level threshold table is established, and the disease risk assessment index is obtained through multivariate regression analysis. In the feature extraction stage, the disease risk assessment index is standardized and the data volume is enhanced. The feature extraction is performed through the convolutional neural network to obtain the initial feature map. The maximum pooling operation is used to reduce the dimension of the feature map, compress the data scale, and retain the key features. The compressed feature data is mapped to the feature vector space through the fully connected layer, and the feature weights are calculated using the attention mechanism to highlight important features. Error backpropagation is performed on the key feature data, and the network parameters are adjusted. Through multiple rounds of iterations, a set of candidate parameters is formed, and finally the parameters with the best performance are selected as disease warning parameters through cross-validation.

[0029] In the grid partitioning process, the disease warning parameters are mapped to the geographic space to form spatial distribution data, and the basic grid units are divided according to the standard of 1 km × 1 km. The terrain characteristics of each grid unit, including altitude, slope, slope aspect, etc., are analyzed to generate terrain partitioning data, calibrate the grid boundaries, and ensure the rationality of the partition. The meteorological elements of the calibrated grid data are clustered, and the areas with similar meteorological characteristics are divided into homogeneous blocks. The data between grids are supplemented by spatial interpolation to form a continuous warning field, and finally the risk level is divided to obtain the partition warning data. In the dynamic optimization stage, the partition warning data is compared and verified with the actual incidence, and the warning accuracy, advance amount and coverage are statistically analyzed to generate the evaluation index value. The indicator threshold is set for judgment, and the warning effect table is obtained by summary analysis. The warning effect table is analyzed for spatiotemporal correlation to find out the key factors affecting the warning accuracy and form the associated feature data. The warning parameters are optimized through precision improvement processing, and the targeted optimization scheme is generated. Finally, the parameters are corrected using the adaptive algorithm to output the final precise targeted warning information of the kiwifruit canker meteorological disaster.

[0030] For example, a kiwi fruit orchard has installed 25 stratified meteorological monitoring stations at different altitudes of 300-800 meters above sea level. Through equipment status detection, it was found that the data of 2 monitoring stations were abnormal and were eliminated. The data of the remaining 23 monitoring stations were evaluated for quality, of which 95% had a data quality score of more than 90 points. Analysis of historical data found that the park had 76 cases of ulcer disease in the past three years, including 45 mild infections, 21 moderate infections, and 10 severe infections. Multivariate statistical analysis determined that the temperature of 15-25°C and relative humidity greater than 85% for more than 48 hours were high-risk thresholds. After deep learning model training, the warning accuracy rate of the disease warning in 2023 reached 87%, and the warning information was issued 72 hours in advance on average. The coverage rate of warning information reached 92%, which is about 30% higher than the accuracy of traditional warning methods.

[0031] In the embodiment of the present application, the temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring station are processed by the Internet of Things in real time transmission, so as to realize the three-dimensional collection and real-time monitoring of multi-level meteorological data, establish a complete three-dimensional meteorological monitoring network data, provide timely and accurate basic data support for disease early warning, and standardize the historical occurrence time, severity and distribution range data of kiwifruit canker disease and the meteorological observation data and cultivation management data of the corresponding time period to establish a systematic historical disease database, which provides reliable historical data for disease law analysis. Support, at the same time, multivariate statistical analysis and processing of three-dimensional meteorological monitoring network data and historical disease databases, extraction of disease risk assessment indicators, accurate identification of key factor combinations that cause diseases, and feature extraction of disease risk assessment indicators through deep learning algorithms to obtain disease warning parameters with predictive capabilities, thereby improving the accuracy and scientificity of warnings. Grid-based zoning of disease warning parameters is then used to achieve spatial refinement of warning information. Finally, the accuracy, lead time and coverage of zoned warning data are dynamically optimized to continuously improve the warning effect, ultimately achieving accurate warning of kiwifruit canker.

[0032] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0033] (1) Detect and process the device version number and operating status of the hierarchical meteorological monitoring station to obtain device status data, and perform real-time monitoring and processing on the device status data to obtain device operating parameters;

[0034] (2) Performing threshold analysis on the device operating parameters to obtain a list of abnormal devices, and removing the devices in the list of abnormal devices to obtain a list of valid devices;

[0035] (3) collecting and processing the temperature, humidity, precipitation, wind direction, wind speed, and light data in layers according to the valid equipment list to obtain a first meteorological data set, and collecting and processing the soil moisture data and canopy temperature and humidity data in layers to obtain a second meteorological data set;

[0036] (4) performing data quality assessment processing on the first meteorological data set and the second meteorological data set to obtain a meteorological data quality score, and performing data cleaning processing on the first meteorological data set and the second meteorological data set according to the meteorological data quality score to obtain cleaned meteorological data;

[0037] (5) completing the cleaned meteorological data by using a time series interpolation algorithm to obtain completed meteorological data, and performing standardized conversion processing on the completed meteorological data to obtain standardized meteorological data;

[0038] (6) The standardized meteorological data are integrated and processed from multiple sources through data fusion algorithms to obtain three-dimensional meteorological monitoring network data.

[0039] Specifically, the basic information of the hierarchical meteorological monitoring station is collected, including the device version number (referring to the model and software version information of the monitoring device) and the operating status (referring to the working status of the device, battery power, signal strength and other parameters). Specifically, when collecting the device version number, the hardware model (such as KQX-500 automatic weather station) and software version number (such as V2.3.1) of the device are recorded; when collecting the operating status, the operating parameters such as the remaining battery percentage, data transmission signal strength (unit: dBm), and device temperature are recorded. This information is collected regularly through the data acquisition module, and the collection frequency is set to once every 5 minutes. The collected data forms the device status data, and then a real-time monitoring mechanism is established to obtain the device operating parameters through data analysis. When analyzing the device operating parameters, multiple monitoring thresholds are set: the battery power threshold is 20%, and it is considered abnormal if it is lower than this value; the signal strength threshold is -90dBm, and it is considered abnormal if it is lower than this value; the data transmission success rate threshold is 95%, and it is considered abnormal if it is lower than this value; the device operating temperature range is -10℃ to 50℃, and it is considered abnormal if it exceeds this range. The device operating parameters are judged through these thresholds to generate a list of abnormal devices, and the device data in the list is eliminated to form a final list of valid devices.

[0040] According to the list of valid equipment, carry out hierarchical data collection. For the first meteorological data set, the collection accuracy of temperature data is 0.1℃, and the collection interval is 5 minutes; the collection accuracy of humidity data is 1%RH, and the collection interval is 5 minutes; the collection accuracy of precipitation data is 0.1mm, and the collection interval is 1 minute; the collection accuracy of wind direction data is 1 degree, and the collection interval is 1 minute; the collection accuracy of wind speed data is 0.1m / s, and the collection interval is 1 minute; the collection accuracy of light data is 1W / ㎡, and the collection interval is 5 minutes. For the second meteorological data set, the collection accuracy of soil moisture data is 0.1%, and the collection interval is 30 minutes; the collection accuracy of canopy temperature and humidity data is 0.1℃ and 1%RH, respectively, and the collection interval is 5 minutes. The data quality assessment adopts a multi-dimensional scoring mechanism, and the quality score of each data point is calculated by four indicators: data integrity (weight 0.3), data accuracy (weight 0.3), data consistency (weight 0.2), and data timeliness (weight 0.2). Data integrity is assessed by the proportion of missing values, data accuracy is assessed by the degree of conformity with the physical reasonable range, data consistency is assessed by the change range of adjacent data points, and data timeliness is assessed by the data transmission delay time. The weighted scores of these four indicators are used as the final meteorological data quality score, and data points with scores below 80 points are eliminated to form cleaned meteorological data.

[0041] The cleaned meteorological data is subjected to time series interpolation processing, using a combination of linear interpolation and spline interpolation. For data with relatively gentle changes (such as temperature and humidity), cubic spline interpolation is used; for data with strong mutation (such as precipitation and wind speed), piecewise linear interpolation is used. After the interpolation is completed, the data is standardized and converted. The minimum-maximum standardization method is used to uniformly convert data of different dimensions to the [0,1] interval to obtain standardized meteorological data. The data fusion algorithm uses a multi-source data fusion method based on the Dempster-Shafer evidence theory to integrate standardized meteorological data. Data from different sources are weighted according to their credibility, and fused through evidence synthesis rules to finally form a three-dimensional meteorological monitoring network data.

[0042] For example, the base has deployed 30 stratified meteorological monitoring stations, distributed at different altitudes of 300-800 meters above sea level. After the equipment status detection, it was found that the battery power of 3 monitoring stations was less than 15%, the signal strength of 2 monitoring stations was less than -95dBm, and the data transmission success rate of 1 monitoring station was 92%. These 6 abnormal devices were eliminated, and 24 valid devices remained. In one data collection cycle, 2880 temperature data (24 stations × 288 points / day), 2880 humidity data, 34560 precipitation data (24 stations × 1440 points / day), 34560 wind direction data, 34560 wind speed data, 2880 light data, 1152 soil moisture data (24 stations × 48 points / day), and 5760 canopy temperature and humidity data (24 stations × 288 points / day × 2 parameters) were collected. After data quality assessment, the average quality score of temperature data was 92.5 points, humidity data was 91.8 points, precipitation data was 88.6 points, wind direction data was 87.9 points, wind speed data was 89.2 points, light data was 90.5 points, soil moisture data was 93.1 points, and canopy temperature and humidity data was 90.8 points. There were 238 data points with a quality score lower than 80 points, accounting for 0.2% of the total data volume. These data were removed and supplemented by interpolation algorithms. After standardized conversion and data fusion, a complete three-dimensional meteorological monitoring network data was finally formed, providing a data basis for subsequent disease early warning analysis.

[0043] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0044] (1) The historical occurrence time data of kiwifruit canker were processed in a unified time series format to obtain standard time series data, and seasonal analysis was performed on the standard time series data to obtain time period characteristics;

[0045] (2) Performing quantitative processing on the disease severity data to obtain disease severity data, and performing spatial distribution processing on the disease severity data to obtain disease distribution characteristics;

[0046] (3) Perform correlation analysis on the disease distribution characteristics and time period characteristics to obtain spatiotemporal correlation indicators, and perform weight distribution on the spatiotemporal correlation indicators to obtain disease characteristic weights;

[0047] (4) Perform data normalization on meteorological observation data to obtain standard meteorological data, and perform missing value filling on standard meteorological data to obtain complete meteorological data;

[0048] (5) performing encoding conversion processing on the cultivation management data to obtain management characteristic data, and performing correlation analysis processing on the management characteristic data to obtain management influencing factors;

[0049] (6) The complete meteorological data, disease characteristic weights, and management influencing factors are integrated through a data fusion algorithm to obtain a historical disease database.

[0050] Specifically, first, the time records in different formats are uniformly converted into a standard format. The collected raw data records are in various formats, including year-month-day hour-minute, year-month-day, lunar calendar records, etc. These data are uniformly converted into the standard format of "YYYY-MM-DD HH:mm:ss" to form standard time series data. Seasonal analysis is performed on the standard time series data. The time series decomposition method is used to decompose the data into trend items, seasonal items, and random items, and the seasonal variation characteristics are extracted to determine the periodic laws of the disease and obtain the time period characteristics. The grade quantification method is used to process the disease severity data, and quantitative standards are established based on indicators such as the proportion of lesion area, lesion depth, and the number of affected branches. It is divided into three levels: mild (lesion area accounts for less than 10%, lesion depth is less than 1mm, and the number of affected branches is less than 5), moderate (lesion area accounts for 10%-30%, lesion depth is 1-3mm, and the number of affected branches is 5-15), and severe (lesion area accounts for more than 30%, lesion depth is greater than 3mm, and the number of affected branches is more than 15), and the values ​​are assigned to 1, 2, and 3 respectively to form the disease grade data. Combined with the latitude and longitude coordinate information of the disease location, the Kriging spatial interpolation method is used to perform spatial distribution analysis on the disease grade data to obtain the disease distribution characteristics.

[0051] The disease distribution characteristics and time period characteristics were correlated and analyzed, and the Pearson correlation coefficient was used to calculate the correlation between spatial distribution and time series, and the spatiotemporal correlation matrix was established. For each spatiotemporal correlation indicator, a weight was assigned according to its impact on the occurrence and development of the disease. The weight assignment adopted the hierarchical analysis method. First, a judgment matrix was constructed, and each indicator was compared two by two. The eigenvalues ​​and eigenvectors were calculated. After the consistency test, the weight coefficients of each indicator were obtained to form the disease characteristic weights. The meteorological observation data were normalized, and the minimum-maximum normalization method was used to uniformly convert the meteorological element data of different dimensions (temperature, humidity, precipitation, wind speed, etc.) to the [0,1] interval for subsequent analysis. Multiple interpolation was used to fill the missing values ​​in the normalized data. Linear interpolation was used for time series data, and inverse distance weighted interpolation was used for spatial distribution data to ensure the integrity and continuity of the data and obtain complete meteorological data.

[0052] Cultivation management data contains a variety of operation information such as fertilization, pruning, and plant protection, which need to be coded and converted. For fertilization data, quantitative coding is performed according to the type of fertilizer, application amount, and application method; for pruning data, coding is performed according to pruning intensity, time, and method; for plant protection data, coding is performed according to the type of agent, concentration, and spraying method. Correlation analysis is performed on the coded management characteristic data, and the correlation coefficient between each management measure and the occurrence of diseases is calculated. The management factors that have a significant impact on the occurrence of diseases are identified, and the management influencing factors are obtained. The complete meteorological data, disease characteristic weights, and management influencing factors are integrated through the data fusion algorithm, and the DS evidence theory is used for multi-source data fusion. First, the basic probability distribution function is constructed to determine the credibility of each data source, and then the DS combination rule is used to synthesize evidence, and finally a unified historical disease database is formed.

[0053] For example, the original onset time records collected include "morning of May 15, 2020", "May 16, 2020", "April 23 of the lunar calendar", etc., which are converted into standard formats such as "2020-05-15 10:00:00" after unified processing. Seasonal analysis found that the ulcer disease in the base mainly occurred in two periods, April-June (spring) and September-October (autumn), with an incidence rate of 65% in spring and 35% in autumn. The incidence data showed that among the 127 recorded outbreaks, there were 78 mild outbreaks, 35 moderate outbreaks, and 14 severe outbreaks. Spatial distribution analysis showed that the slope area with an altitude of 500-600 meters had the highest incidence frequency, accounting for 45% of the total number of outbreaks. The spatiotemporal correlation analysis showed that the correlation coefficient between spring outbreaks and temperature was 0.82, and the correlation coefficient with humidity was 0.75. After normalization of meteorological data, about 3% of missing values ​​were filled by multiple interpolation. Analysis of cultivation management data showed that the disease peaked 7-10 days after pruning, and the incidence increased by about 20% after the application of nitrogen-containing fertilizers. Finally, through data fusion, a complete historical disease database including meteorological conditions, disease characteristics and management measures was established to provide data support for the establishment of subsequent early warning models.

[0054] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0055] (1) Perform periodic analysis and processing on the temperature data in the three-dimensional meteorological monitoring network data to obtain the temperature change law, and divide the temperature change law into threshold intervals to obtain the temperature risk index;

[0056] (2) Performing statistical processing on the duration of humidity data in the three-dimensional meteorological monitoring network data to obtain humidity persistence characteristics, and then performing grade assessment processing on the humidity persistence characteristics to obtain the humidity risk index;

[0057] (3) Conduct coupling analysis on the temperature risk index and the humidity risk index to obtain the environmental combination index, and perform threshold verification on the environmental combination index to obtain the key environmental factors;

[0058] (4) Perform frequency statistical processing on the disease patterns in the historical disease database to obtain the characteristics of the high-incidence period, and match the characteristics of the high-incidence period with key environmental factors to obtain risk prediction factors;

[0059] (5) Performing weight calculation processing on the risk prediction factors to obtain indicator weight data, and performing graded value assignment processing on the indicator weight data to obtain a grade threshold table;

[0060] (6) The level threshold table is comprehensively calculated and processed through multivariate regression analysis to obtain the disease risk assessment index.

[0061] Specifically, a periodic analysis is conducted, and the time series decomposition method is used to decompose the temperature data into three scales: annual, monthly, and daily. The annual scale focuses on the seasonal variation of temperature, the monthly scale focuses on the fluctuation characteristics of the monthly average temperature, and the daily scale focuses on the variation characteristics of the daily temperature difference. The periodic information of temperature variation is extracted by Fourier transform, and the main periodic components are determined. According to the analysis results, the temperature variation law is divided into different risk intervals: the temperature below 10℃ is a low risk interval, assigned a value of 0.2; the temperature between 10-15℃ or 25-30℃ is a medium risk interval, assigned a value of 0.5; the temperature between 15-25℃ is a high risk interval, assigned a value of 0.8; the temperature above 30℃ is an extremely high risk interval, assigned a value of 1.0, and the temperature risk index is obtained. The humidity data processing focuses on the duration characteristics of relative humidity, and the duration of relative humidity exceeding a specific threshold is counted. The humidity data of each observation point is analyzed for continuity, and the number of hours when the relative humidity exceeds 75%, 85%, and 95% is recorded. The duration characteristics are divided into levels according to the degree of hazard: when the relative humidity exceeds 75% for less than 24 hours, it is low risk and is assigned a value of 0.3; when the relative humidity exceeds 85% for 24-48 hours, it is medium risk and is assigned a value of 0.6; when the relative humidity exceeds 95% for more than 48 hours, it is high risk and is assigned a value of 0.9, forming a humidity risk index.

[0062] The coupling analysis of temperature risk index and humidity risk index adopts weighted product method, considering the synergistic effect of the two factors. The temperature risk index is multiplied by the humidity risk index, and the weight coefficient is set according to the historical data analysis results. The temperature weight is 0.6 and the humidity weight is 0.4. The environmental combination index is calculated. The threshold of the environmental combination index is verified, and the index value is divided into four levels: 0-0.3 is low risk, 0.3-0.6 is medium risk, 0.6-0.8 is high risk, and 0.8-1.0 is extremely high risk. The key environmental factors are determined through verification. The incidence records are extracted from the historical disease database, and the incidence frequency is counted according to the time dimension. The statistical period is set to ten days (10 days), and the proportion of the number of incidences in each ten days to the total number of incidences in the whole year is calculated to determine the high incidence period. The period with an incidence frequency greater than 10% is defined as the high incidence period, and the corresponding temperature and humidity characteristics are recorded to obtain the characteristics of the high incidence period. The characteristics of the high incidence period are matched and analyzed with the key environmental factors, and a corresponding relationship matrix is ​​established. The factor combination with significant correlation is extracted to form a risk prediction factor.

[0063] The entropy weight method is used to calculate the weight of risk prediction factors. First, each prediction factor is normalized, and the information entropy of each factor in different periods is calculated. The weight is determined according to the size of the information entropy to obtain the indicator weight data. The indicator weight data is divided into five levels according to the risk level: very low risk (0-0.2), low risk (0.2-0.4), medium risk (0.4-0.6), high risk (0.6-0.8), and very high risk (0.8-1.0), and a level threshold table is established. A risk assessment model is established through multivariate regression analysis, with the probability of disease as the dependent variable and each prediction factor as the independent variable. The stepwise regression method is used to screen the variables with a significance level of 0.05, and the regression equation is established. The regression coefficient is tested for significance and model diagnosis is performed to ensure the reliability of the model, and finally the disease risk assessment index is obtained.

[0064] For example, through the analysis of temperature data from 2021 to 2023, it was found that the frequency of ulcer disease was highest when the daily average temperature was 15-25℃ and the daily temperature difference was 8-12℃. From April 15 to April 25, 2022, the daily average temperature was stable at 18-22℃, the daily temperature difference was maintained at about 10℃, and the temperature risk index reached 0.8. During the same period, the relative humidity remained above 90% for 72 hours, and the humidity risk index was 0.9. The temperature and humidity risk indexes were coupled and analyzed, and the environmental combination index was calculated to be 0.85, which is in an extremely high risk range. Historical data statistics show that ulcer disease in the park mainly occurs in spring from April to May and autumn from September to October, among which the frequency of disease is highest in late April, accounting for 25% of the total number of disease cases throughout the year. Matching the characteristics of the high incidence period with environmental factors, it is found that the temperature of 15-22℃ and the relative humidity greater than 90% for more than 48 hours are the key conditions for causing the disease. After weight calculation and graded assignment, the weight of temperature duration is 0.35, the weight of humidity duration is 0.40, and the weight of daily temperature difference is 0.25. Finally, through multiple regression analysis, a risk assessment index system including temperature, humidity, duration and other factors was established. The accuracy rate of ulcer disease warning in 2023 reached more than 85%.

[0065] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0066] (1) Performing feature standardization processing on the disease risk assessment indicators to obtain standardized indicator data, and performing data enhancement processing on the standardized indicator data to obtain a training data set;

[0067] (2) Performing convolutional layer feature extraction processing on the training data set to obtain an initial feature map, and performing pooling layer dimensionality reduction processing on the initial feature map to obtain compressed feature data;

[0068] (3) Performing full-connection layer mapping processing on the compressed feature data to obtain a feature vector, and performing weight calculation processing on the feature vector through the attention mechanism to obtain key feature data;

[0069] (4) performing error back propagation processing on the key feature data to obtain a parameter adjustment scheme, and iteratively updating the parameter adjustment scheme to obtain an alternative parameter set;

[0070] (5) The performance of the candidate parameter sets is evaluated through cross-validation to obtain disease warning parameters.

[0071] Specifically, the disease risk assessment indicators are first subjected to feature standardization, and the minimum-maximum standardization method is used to unify the indicator data of different dimensions into the [0,1] interval. The standardization process normalizes multi-dimensional data such as temperature indicators, humidity indicators, and wind speed indicators to eliminate the influence of dimensions. The standardized data is subjected to data enhancement processing, and the sliding window method is used to generate time series features. The window size is set to 24 hours and the step length is 1 hour. Each window contains multiple feature values ​​such as temperature and humidity to form a high-dimensional training data set. At the same time, the amount of data is expanded by adding Gaussian noise, random scaling, and other methods to improve the generalization ability of the model. The training data set is input into the convolutional neural network for feature extraction, and the convolution layer uses multiple convolution kernels of different sizes to scan the input data. The first layer of convolution uses 32 3×3 convolution kernels to extract local features; the second layer of convolution uses 64 5×5 convolution kernels to extract higher-level feature combinations. Each convolution operation is followed by a ReLU activation function to enhance the nonlinear expression ability of the feature and obtain the initial feature map. The initial feature map is subjected to a maximum pooling operation with a pooling window size of 2×2 and a step size of 2. The feature dimension is reduced by taking the maximum value of the local area while retaining the significant features to obtain compressed feature data.

[0072] The compressed feature data is mapped through a fully connected layer. The first fully connected layer contains 256 neurons, and the second fully connected layer contains 128 neurons. The dropout mechanism is used to prevent overfitting, and the dropout rate is set to 0.5. After mapping through the fully connected layer, the feature vector is obtained, and then the attention mechanism is introduced to calculate the attention weights for each dimension in the feature vector. The attention mechanism weights the features through a 128-dimensional attention vector, highlights the contribution of important features, suppresses the influence of minor features, and finally obtains the key feature data. Error backpropagation is performed on the key feature data, the mean square error between the predicted value and the true value is calculated, and the network parameters are optimized using the gradient descent method. The learning rate is initially set to 0.001, and the Adam optimizer is used for parameter update. After every 100 iterations, the learning rate is reduced to 0.9 times the original to ensure the stability of training. Through multiple rounds of iterative training, the network weights and bias parameters are continuously adjusted, and multiple sets of candidate parameter sets are accumulated.

[0073] The performance of the candidate parameter sets was evaluated using 10-fold cross validation. The data set was randomly divided into 10 parts, 9 of which were used as training sets and 1 as validation sets in turn. The three evaluation indicators of prediction accuracy, recall rate and F1 score were calculated for each set of parameters, and the parameter set with the best comprehensive performance was selected as the final disease warning parameter.

[0074] For example, the disease risk assessment indicator data for 2021-2023 were collected, including 20 feature dimensions such as temperature, humidity, and wind speed, with an original data volume of 26,280 (one record per hour × 24 hours × 365 days × 3 years). After standardization, all indicator values ​​are mapped to the [0,1] interval, such as the original temperature data of 5-35℃ is converted to a standardized value of 0-1. Through sliding window and data enhancement processing, the training data set was expanded to 78,840. After input into the convolutional neural network, the first layer of convolution produces 32 feature maps, and the second layer of convolution produces 64 feature maps. After the pooling layer is processed, the feature dimension is reduced by 75%. The fully connected layer reduces the feature dimension from 1024 to 128. The attention mechanism calculation shows that the weights of the two features, temperature duration and humidity accumulation value, are 0.25 and 0.22, respectively, which are significantly higher than other features. After 100 rounds of iterative training, 5 sets of alternative parameter sets are obtained. Through cross-validation, the optimal parameter group achieved a prediction accuracy of 87.5% on the validation set, predicting the risk of disease occurrence 72 hours in advance on average, and keeping the false alarm rate within 10%. The disease warning parameters finally selected were applied to the 2023 spring warning work, successfully warning of four major disease outbreaks, winning sufficient time for prevention and control work.

[0075] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0076] (1) Performing geographic spatial mapping on disease warning parameters to obtain spatial distribution data, and then dividing the spatial distribution data into equal intervals to obtain basic grid units;

[0077] (2) Perform terrain feature analysis on the basic grid cells to obtain terrain partition data, and perform boundary calibration on the terrain partition data to obtain calibration grid data;

[0078] (3) Perform meteorological element clustering on the calibration grid data to obtain homogeneous blocks, and then perform spatial interpolation on the homogeneous blocks to obtain continuous warning fields;

[0079] (4) Classify the risk levels of continuous warning sites and obtain regional warning data.

[0080] Specifically, the longitude and latitude coordinate information of each monitoring point is associated with the warning parameter value, and the longitude and latitude are accurate to 6 decimal places to ensure that the positioning accuracy is within the meter range. The spatial interpolation method is used to continuously process the discrete monitoring point data to generate spatial distribution data covering the entire planting area. Subsequently, the spatial distribution data is divided into equal intervals according to the standard grid of 1 km × 1 km. Each grid unit records the statistical characteristics such as the average value, maximum value, and minimum value of the warning parameters in the area to form a basic grid unit. The terrain feature analysis of the basic grid unit includes three dimensions: altitude, slope, and slope direction. The altitude data is extracted from the digital elevation model with an accuracy of 1 meter; the slope calculation adopts the elevation difference method of adjacent grid cells and is divided into four categories: flat land (0-5 degrees), gentle slope (5-15 degrees), steep slope (15-25 degrees) and dangerous slope (greater than 25 degrees); the slope direction is divided into four main directions of east, south, west, and north and four secondary directions of southeast, southwest, northwest, and northeast. These terrain features are comprehensively analyzed to obtain terrain zoning data. The boundaries of the terrain partition data are calibrated, mainly by smoothing the transition zones between adjacent partitions to avoid sudden changes, and finally the calibrated grid data is obtained.

[0081] The meteorological element clustering analysis of the calibration grid data uses the K-means clustering algorithm to group grid cells with similar meteorological characteristics. The clustering indicators include key meteorological elements such as daily average temperature, daily temperature difference, relative humidity, and wind speed, while considering the influence weight of terrain factors. By calculating the Euclidean distance between each grid cell, the optimal number of clusters is determined, and the entire warning area is divided into several homogeneous blocks with similar meteorological characteristics. The warning data in the homogeneous blocks are processed by Kriging spatial interpolation, and the interpolation grid size is set to 100 meters × 100 meters to generate a high-resolution continuous warning field. The risk level of the continuous warning field is divided into four levels according to the statistical distribution characteristics of the warning value: low risk (warning value 0-0.25), medium risk (warning value 0.25-0.5), high risk (warning value 0.5-0.75) and extremely high risk (warning value 0.75-1.0). The area statistics and boundary extraction of each risk level area are performed to form the partitioned warning data, which provides a spatial reference for the subsequent accurate warning information release.

[0082] For example, the base has a total area of ​​500 hectares, an altitude span of 300-800 meters, and 25 meteorological monitoring points are deployed. First, the warning parameter values ​​of these monitoring points are spatially mapped, and the coordinates of the monitoring points are accurate to 6 decimal places, such as (26.123456°N, 104.123456°E). According to the standard of 1 km × 1 km, 36 basic grid units are divided, and each grid unit contains 0-3 monitoring points. The terrain feature analysis shows that the base contains 120 hectares of flat areas (slope 0-5 degrees), 200 hectares of gentle slope areas (slope 5-15 degrees), 150 hectares of steep slope areas (slope 15-25 degrees), and 30 hectares of dangerous slope areas (slope greater than 25 degrees). The main slope directions are southeast and southwest. Through K-means cluster analysis, the 36 grid cells were divided into 5 homogeneous blocks: Block A (flat area, 160 hectares), Block B (southeast gentle slope area, 120 hectares), Block C (southwest gentle slope area, 100 hectares), Block D (eastern steep slope area, 80 hectares), Block E (western steep slope area, 40 hectares). Spatial interpolation was performed on these 5 blocks to generate 5,000 100m×100m high-precision early warning grids. The final risk level classification results showed that the low-risk area accounted for 15% (75 hectares) of the total area, mainly distributed in the flat area; the medium-risk area accounted for 35% (175 hectares), mainly distributed in the gentle slope area; the high-risk area accounted for 40% (200 hectares), mainly distributed in the steep slope area with southeast and southwest slopes; the extremely high-risk area accounted for 10% (50 hectares), mainly distributed in areas with complex terrain and special microclimate. The regional early warning data realizes the precise spatial positioning of disease risks and provides a scientific basis for the formulation of differentiated prevention and control measures.

[0083] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0084] (1) Compare the regional warning data with the actual incidence data to obtain verification result data, and perform statistical analysis on the verification result data to obtain evaluation index values;

[0085] (2) Perform threshold judgment processing on the evaluation index values ​​to obtain the index compliance status, and classify and summarize the index compliance status to obtain an early warning effect table;

[0086] (3) Performing spatiotemporal correlation analysis on the warning effect table to obtain related feature data, and performing precision improvement processing on the related feature data to obtain a targeted optimization solution;

[0087] (4) The parameters of the targeted optimization scheme were corrected through an adaptive algorithm to obtain accurate and targeted early warning information on kiwifruit canker meteorological disasters.

[0088] Specifically, a spatiotemporal matching relationship is established, and the spatial location (longitude and latitude coordinates) and time point (accurate to the hour) of the warning data are matched one by one with the actual onset record. The time difference between the time when the warning signal is issued and the actual onset time in each warning grid is counted to calculate the warning lead time; at the same time, the accuracy of the warning intensity is evaluated by comparing the warning level with the actual degree of onset; the spatial overlap between the warning range and the actual onset area is compared to evaluate the spatial accuracy of the warning. These comparison results are recorded as verification result data, and then statistical analysis is performed to calculate three core evaluation indicators: warning accuracy (the proportion of correct warning times to total warning times), warning lead time (average warning lead time) and spatial coverage (the overlap between the warning area and the actual onset area). The threshold judgment of the evaluation index value adopts a grading standard: the threshold of the warning accuracy is set at 85%, and it is judged as not meeting the standard if it is lower than this value; the threshold of the warning lead time is set at 48 hours, and it is judged as not meeting the standard if it is lower than this value; the threshold of the spatial coverage is set at 80%, and it is judged as not meeting the standard if it is lower than this value. The three indicators were combined and analyzed to establish a four-level evaluation system: A (all three indicators met), B (two indicators met), C (one indicator met), and D (no indicator met). The early warning effects in different regions and periods were classified and summarized according to these four levels to form an early warning effect table, which includes the rating results and specific indicator values ​​of each early warning grid.

[0089] The spatiotemporal correlation analysis of the early warning effect table focuses on the correlation between the early warning effect and factors such as terrain characteristics, meteorological conditions, and planting density. Through correlation analysis, the key factors affecting the accuracy of early warning are found, such as the correlation coefficient between terrain undulation and early warning accuracy, the correlation coefficient between plant density and spatial coverage, and the correlation coefficient between microclimate characteristics and early warning advance, forming correlation feature data. The correlation feature data is processed for accuracy improvement, and the threshold of early warning parameters is adjusted for different characteristic value intervals, the early warning criteria are optimized, and a targeted optimization scheme is established. The adaptive algorithm dynamically corrects the early warning parameters based on the targeted optimization scheme. The algorithm adopts an online learning method to receive new observation data and disease records in real time and continuously update the early warning parameters. For areas with poor early warning effects, the weight of environmental factors is increased; for periods with frequent missed reports, the early warning threshold is lowered; for areas with concentrated false reports, the strictness of the early warning conditions is increased. Through this adaptive adjustment, the early warning parameters are continuously optimized, and finally more accurate kiwi canker meteorological disaster early warning information is output.

[0090] For example, a base with an area of ​​500 hectares is divided into 50 early warning grids. In these three months, the early warning system issued 85 early warning signals, and ulcer disease actually occurred 62 times. Through comparative analysis, 54 accurate warnings, 8 false reports, and 6 missed reports were issued, with an early warning accuracy rate of 87.1%. The early warning time distribution shows that there are 25 early warnings 72 hours in advance, 20 early warnings 48-72 hours in advance, 15 early warnings 24-48 hours in advance, and 2 early warnings within 24 hours in advance, with an average early warning time of 56.5 hours. The spatial distribution comparison shows that the spatial coverage rate of 40 early warning grids exceeds 85%, 8 grids are between 75%-85%, and 2 grids are below 75%, with an average spatial coverage rate of 83.6%. According to the evaluation index threshold, 31 grids meet the A-level standard (all three indicators meet the standard), 15 grids are B-level (two indicators meet the standard), and 4 grids are C-level (one indicator meets the standard). The spatiotemporal correlation analysis found that the warning effect was best in the gentle slope area with an altitude of 500-600 meters, with an accuracy rate of 92%; in areas with plant density greater than 2,000 plants per hectare, the spatial coverage rate generally exceeded 90%; and the average advance time for warning on the southeast slope reached 65 hours. Based on these analysis results, differentiated parameter optimization schemes were adopted for different regions: for areas with complex terrain, the weight of meteorological factors was increased by 20%; for high-density planting areas, the spatial resolution was increased to 50 meters × 50 meters; for special microclimate areas, the monitoring frequency was increased at the hourly level. After the adaptive algorithm was adjusted, in the subsequent one-month verification, the overall warning accuracy rate was increased to 91.5%, the average advance time was increased to 68 hours, and the spatial coverage rate reached 88.2%.

[0091] The above describes the precise targeted early warning method for meteorological disasters caused by kiwifruit canker in the embodiment of the present application. The following describes the precise targeted early warning device for meteorological disasters caused by kiwifruit canker in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the precise targeted early warning device for meteorological disasters of kiwifruit canker disease includes:

[0092] Transmission module 201 is used to perform real-time Internet of Things transmission processing on the temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring station to obtain three-dimensional meteorological monitoring network data;

[0093] Processing module 202 is used to standardize the historical occurrence time, disease severity, distribution range data of kiwifruit canker disease and the meteorological observation data and cultivation management data of the corresponding period to obtain a historical disease database;

[0094] The analysis module 203 is used to perform multivariate statistical analysis on the three-dimensional meteorological monitoring network data and the historical disease database to obtain disease risk assessment indicators;

[0095] An extraction module 204 is used to perform feature extraction processing on the disease risk assessment index through a deep learning algorithm to obtain disease warning parameters;

[0096] A partitioning module 205 is used to perform grid partitioning processing on the disease warning parameters to obtain partition warning data;

[0097] The optimization module 206 is used to dynamically optimize the accuracy, advance time and coverage of the sub-area warning data to obtain accurate and targeted warning information on the meteorological disaster of kiwifruit canker.

[0098] Through the coordinated cooperation of the above-mentioned components, the temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring stations are processed in real time through the Internet of Things, and the three-dimensional collection and real-time monitoring of multi-level meteorological data are realized. A complete three-dimensional meteorological monitoring network data is established to provide timely and accurate basic data support for disease early warning. By standardizing the historical occurrence time, severity, distribution range data of kiwifruit canker disease and the meteorological observation data and cultivation management data of the corresponding time period, a systematic historical disease database is established to provide reliable support for disease law analysis. With the support of historical data, multivariate statistical analysis and processing are carried out on the three-dimensional meteorological monitoring network data and the historical disease database to extract disease risk assessment indicators, accurately identify the key factor combination that causes the disease, and use deep learning algorithms to perform feature extraction on the disease risk assessment indicators to obtain disease warning parameters with predictive capabilities, thereby improving the accuracy and scientificity of the warning. The disease warning parameters are then processed by gridding and zoning to achieve spatial refinement of the warning information. Finally, the accuracy, lead time and coverage of the zoned warning data are dynamically optimized to continuously improve the warning effect, ultimately achieving accurate warning of kiwifruit canker.

[0099] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the method for precise targeted early warning of meteorological disasters caused by kiwifruit canker.

[0100] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0101] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0102] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A precise targeted early warning method for meteorological disasters of kiwifruit canker, characterized in that: The precise targeted early warning method for meteorological disasters of kiwifruit canker disease comprises: The temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring stations are processed through real-time Internet of Things transmission to obtain three-dimensional meteorological monitoring network data; The historical occurrence time, severity, and distribution range of kiwifruit canker disease, as well as the meteorological observation data and cultivation management data of the corresponding period were standardized to obtain a historical disease database. Performing multivariate statistical analysis on the three-dimensional meteorological monitoring network data and the historical disease database to obtain disease risk assessment indicators; Performing feature extraction processing on the disease risk assessment index through a deep learning algorithm to obtain disease warning parameters; Performing grid-based partitioning processing on the disease warning parameters to obtain partition warning data; The accuracy, lead time and coverage of the zoning warning data are dynamically optimized to obtain precise and targeted warning information on the kiwifruit canker meteorological disaster.

2. The precise targeted early warning method for meteorological disasters of kiwifruit canker according to claim 1, characterized in that: The temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring station are processed by real-time Internet of Things transmission to obtain three-dimensional meteorological monitoring network data, including: Detect and process the device version number and operating status of the hierarchical meteorological monitoring station to obtain device status data, and perform real-time monitoring and processing on the device status data to obtain device operating parameters; Perform threshold analysis on the equipment operating parameters to obtain a list of abnormal equipment, and remove the equipment in the list of abnormal equipment to obtain a list of valid equipment; According to the valid equipment list, the temperature, humidity, precipitation, wind direction, wind speed, and light data are collected and processed in layers to obtain a first meteorological data set, and the soil moisture data and canopy temperature and humidity data are collected and processed in layers to obtain a second meteorological data set; Performing data quality assessment processing on the first meteorological data set and the second meteorological data set to obtain meteorological data quality scores, and performing data cleaning processing on the first meteorological data set and the second meteorological data set according to the meteorological data quality scores to obtain cleaned meteorological data; The cleaned meteorological data is supplemented by a time series interpolation algorithm to obtain supplemented meteorological data, and the supplemented meteorological data is standardized and converted to obtain standardized meteorological data; The standardized meteorological data is integrated and processed from multiple sources through data fusion algorithm to obtain three-dimensional meteorological monitoring network data.

3. The precise targeted early warning method for meteorological disasters of kiwifruit canker according to claim 1, characterized in that: The historical occurrence time, disease severity, distribution range data of kiwifruit canker disease and the meteorological observation data and cultivation management data of the corresponding period are standardized to obtain a historical disease database, including: The historical occurrence time data of kiwifruit canker were processed in a unified time series format to obtain standard time series data, and seasonal analysis was performed on the standard time series data to obtain time period characteristics. Carry out grade quantification processing on the disease severity data to obtain disease severity data, and carry out spatial distribution processing on the disease severity data to obtain disease distribution characteristics; Performing correlation analysis on the disease distribution characteristics and the time period characteristics to obtain time-space correlation indicators, and performing weight distribution processing on the time-space correlation indicators to obtain disease characteristic weights; Normalize the meteorological observation data to obtain standard meteorological data, and fill in the missing values ​​of the standard meteorological data to obtain complete meteorological data; Performing coding conversion processing on the cultivation management data to obtain management characteristic data, and performing correlation analysis processing on the management characteristic data to obtain management influencing factors; The complete meteorological data, the disease characteristic weights and the management impact factors are integrated through a data fusion algorithm to obtain a historical disease database.

4. The precise targeted early warning method for meteorological disasters of kiwifruit canker according to claim 1, characterized in that: The multivariate statistical analysis and processing of the three-dimensional meteorological monitoring network data and the historical disease database is performed to obtain disease risk assessment indicators, including: Performing periodic analysis and processing on the temperature data in the three-dimensional meteorological monitoring network data to obtain a temperature variation law, and performing threshold interval division processing on the temperature variation law to obtain a temperature risk index; Performing duration statistics processing on humidity data in the three-dimensional meteorological monitoring network data to obtain humidity persistence characteristics, and performing grade assessment processing on the humidity persistence characteristics to obtain a humidity risk index; Performing coupling analysis processing on the temperature risk index and the humidity risk index to obtain an environmental combination index, and performing threshold verification processing on the environmental combination index to obtain a key environmental factor; Performing frequency statistics on the disease patterns in the historical disease database to obtain high-incidence period characteristics, and matching the high-incidence period characteristics with the key environmental factors to obtain risk prediction factors; Performing weight calculation processing on the risk prediction factors to obtain indicator weight data, and performing graded value assignment processing on the indicator weight data to obtain a grade threshold table; The grade threshold table is comprehensively calculated and processed through multivariate regression analysis to obtain a disease risk assessment index.

5. The precise targeted early warning method for meteorological disasters of kiwifruit canker according to claim 1, characterized in that: The feature extraction process of the disease risk assessment index is performed by the deep learning algorithm to obtain the disease warning parameters, including: Performing feature standardization processing on the disease risk assessment index to obtain standardized index data, and performing data enhancement processing on the standardized index data to obtain a training data set; Performing convolutional layer feature extraction processing on the training data set to obtain an initial feature map, and performing pooling layer dimensionality reduction processing on the initial feature map to obtain compressed feature data; Performing a fully connected layer mapping process on the compressed feature data to obtain a feature vector, and performing a weight calculation process on the feature vector through an attention mechanism to obtain key feature data; Performing error back propagation processing on the key feature data to obtain a parameter adjustment scheme, and performing iterative update processing on the parameter adjustment scheme to obtain an alternative parameter set; The performance of the candidate parameter set is evaluated through cross-validation to obtain disease warning parameters.

6. The precise targeted early warning method for meteorological disasters of kiwifruit canker according to claim 1, characterized in that: The grid-based partitioning process of the disease warning parameters to obtain partition warning data includes: Performing geographic spatial mapping processing on the disease warning parameters to obtain spatial distribution data, and performing equal-interval division processing on the spatial distribution data to obtain basic grid units; Performing terrain feature analysis processing on the basic grid unit to obtain terrain partition data, and performing boundary calibration processing on the terrain partition data to obtain calibration grid data; Performing meteorological element clustering processing on the calibration grid data to obtain homogeneous blocks, and performing spatial interpolation processing on the homogeneous blocks to obtain continuous warning fields; The continuous warning field is divided into risk levels to obtain zone warning data.

7. The precise targeted early warning method for meteorological disasters of kiwifruit canker according to claim 1, characterized in that: The accuracy, advance amount and coverage of the sub-area warning data are dynamically optimized to obtain precise and targeted warning information on kiwifruit canker meteorological disasters, including: Comparing the sub-area warning data with the actual incidence data to obtain verification result data, and statistically analyzing the verification result data to obtain evaluation index values; Performing threshold judgment processing on the evaluation index value to obtain the index compliance status, and classifying and summarizing the index compliance status to obtain an early warning effect table; Performing spatiotemporal correlation analysis on the warning effect table to obtain associated feature data, and performing precision improvement processing on the associated feature data to obtain a targeted optimization solution; The parameters of the targeted optimization scheme are corrected by an adaptive algorithm to obtain accurate targeted early warning information on meteorological disasters caused by kiwifruit canker.

8. A precise targeted early warning device for meteorological disasters caused by kiwifruit canker, used to implement the precise targeted early warning method for meteorological disasters caused by kiwifruit canker as described in any one of claims 1 to 7, characterized in that: The kiwifruit canker meteorological disaster precise targeted early warning device comprises: The transmission module is used to perform real-time Internet of Things transmission processing on the temperature, humidity, precipitation, wind direction, wind speed, light data, soil moisture data and canopy temperature and humidity data collected by the layered meteorological monitoring station to obtain three-dimensional meteorological monitoring network data; The processing module is used to standardize the historical occurrence time, disease severity, distribution range data of kiwifruit canker disease, as well as the meteorological observation data and cultivation management data of the corresponding period to obtain a historical disease database; An analysis module, used to perform multivariate statistical analysis on the three-dimensional meteorological monitoring network data and the historical disease database to obtain disease risk assessment indicators; An extraction module is used to perform feature extraction processing on the disease risk assessment index through a deep learning algorithm to obtain disease warning parameters; A partitioning module is used to perform grid partitioning processing on the disease warning parameters to obtain partition warning data; The optimization module is used to dynamically optimize the accuracy, lead time and coverage of the sub-area warning data to obtain precise and targeted warning information on kiwifruit canker meteorological disasters.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the precise targeted early warning method for meteorological disasters caused by kiwifruit canker disease as described in any one of claims 1 to 7 is implemented.

Images