Method for predicting and early warning of black spot disease of panax ginseng gray mold based on meteorological factors

Abstract

This invention relates to the field of agricultural disease early warning technology, and discloses a method for predicting and warning the risk of ginseng gray mold and black spot based on meteorological factors. The method includes: acquiring disease severity index data, meteorological factor data, and warning thresholds; determining the disease severity index in mid-August; calculating the Pearson correlation coefficient with meteorological factors and screening key meteorological factors to construct original training samples; superimposing random noise onto the original samples to generate enhanced training samples, and merging them to form an expanded training set; dividing the dataset and training multiple candidate prediction models, selecting the optimal prediction model with the highest determination coefficient; extracting features based on meteorological factors of the area to be predicted, inputting them into the optimal model to generate predicted disease severity index values ​​and warning information. Through data augmentation and automatic multi-model selection mechanisms, this invention alleviates the problem of overfitting in small field samples, improves the accuracy and robustness of disease prediction, and provides scientific decision support for the early identification and precise application of pesticides for ginseng diseases.

Description

Technical Field

[0001] This invention relates to the field of agricultural disease early warning technology, specifically a method for predicting and warning the risk of ginseng gray mold and black spot based on meteorological factors. Background Technology

[0002] Ginseng, a traditional and precious Chinese medicinal herb and an important economic crop, plays a vital role in the industry's economic development. In recent years, with the expansion of ginseng cultivation and the intensification of continuous cropping obstacles, black spot and gray mold have become two major foliar diseases restricting the sustainable development of the ginseng industry, often causing large-scale leaf death and a severe decline in root yield and quality. The occurrence of ginseng black spot and gray mold is closely related to field meteorological conditions; specific temperatures, continuous rainfall, and high humidity easily induce the rapid spread of these diseases. However, traditional ginseng disease management relies heavily on manual field observation and post-disease application of pesticides. This approach suffers from delayed early warning, strong subjectivity, and passive control, making it difficult to meet the management needs of large-scale and standardized cultivation. Furthermore, the extensive use of chemical pesticides increases the burden on the ecological environment.

[0003] Existing plant disease early warning systems largely limit their functionality to information retrieval and browsing, exhibiting weak interactivity. They typically rely on single mathematical models to describe disease prevalence patterns, failing to comprehensively capture the complex occurrence and development processes of diseases under different natural environments. Regarding core prediction algorithms, current technologies mostly employ traditional calculation methods such as multiple regression, which are insufficiently capable of fitting the complex nonlinear relationships between meteorological factors and disease severity. Under conditions of small sample data in actual fields, overfitting is highly likely, resulting in low prediction accuracy and robustness. Consequently, these systems cannot provide scientific and reliable data support for the early identification and precise application of pesticides for ginseng diseases. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a risk prediction and early warning method for ginseng gray mold and black spot based on meteorological factors. The technical problem solved by this invention is that traditional ginseng disease management relies heavily on manual field observation and post-disease application, which leads to problems of delayed early warning and subjective judgment. Existing plant disease early warning systems mostly rely on single mathematical models, which are difficult to fully capture the complex development dynamics of diseases. Moreover, existing predictions mostly use traditional methods such as multiple regression, which are insufficient in fitting the nonlinear relationship between meteorological factors and disease occurrence. Under the condition of small field sample data, overfitting is very likely to occur, resulting in low prediction accuracy and robustness.

[0005] To address the above problems, the present invention provides the following technical solution: This invention provides a method for predicting and warning the risk of ginseng gray mold and black spot based on meteorological factors, including: Acquire ginseng black spot disease disease index data, ginseng gray mold disease disease index data, meteorological factor data, and preset early warning thresholds; The disease index for mid-August was determined from the ginseng black spot disease disease index data and the ginseng gray mold disease disease index data, respectively. Calculate the Pearson correlation coefficient between the mid-August disease index and the meteorological factor data respectively, and select the top five in absolute value from high to low to form the key meteorological factors; The key meteorological factors are used as feature values, and the mid-August disease index is used as the target value to form the original training sample. Random noise is superimposed on the original training samples to generate enhanced training samples. The original training samples and the enhanced training samples are merged to form an expanded training set. The expanded training set is divided into a training part and a validation part. Candidate prediction models are trained, and the candidate prediction models with the highest determination coefficients in the validation part are determined as the optimal prediction models. Receive meteorological factor data of the area to be predicted, extract corresponding feature data from the meteorological factor data according to the key meteorological factors and input them into the optimal prediction model to generate a disease index prediction value, and generate a warning information according to the preset warning threshold.

[0006] Furthermore, data on ginseng black spot disease index, ginseng gray mold disease index, meteorological factor data, and preset early warning thresholds are obtained, including: Obtain the ginseng black spot disease disease index data and the ginseng gray mold disease disease index data; Receive survey leaf data generated using the five-point field sampling method; Obtain the whole leaf area and lesion area of ​​a single surveyed leaf from the surveyed leaf data, and determine the percentage value of lesion area and disease level; When the percentage values ​​of the lesion area are 0, greater than 0 and not exceeding 1.0%, 1.1% to 10.0%, 10.1% to 25.0%, 25.1% to 50.0%, and greater than 50.0%, the disease grades are 0, 1, 3, 5, 7, and 9, respectively. The number of leaves surveyed at each level and the total number of leaves surveyed were counted to generate time series data of ginseng black spot disease index and ginseng gray mold disease index, respectively. The time-series data of ginseng black spot disease index and the time-series data of ginseng gray mold disease index both include survey nodes consisting of disease name, location identifier, year, survey date, survey period, and disease index; Survey nodes with a total number of leaves of 0 or missing are marked as invalid nodes, and survey nodes with a total number of leaves greater than 0 are marked as valid survey nodes.

[0007] Furthermore, after generating time-series data on the ginseng black spot disease index and the ginseng gray mold disease index, the following is also included: The survey nodes in the time series data of ginseng black spot disease index and the time series data of ginseng gray mold disease index are sorted according to the survey date, and the first and last survey nodes after sorting are determined as the start point and the end point, respectively. When the invalid node is neither located at the starting point nor at the ending point, and both the preceding and following survey nodes of the invalid node are valid survey nodes, the invalid node is completed using linear interpolation according to the survey dates of the preceding and following survey nodes. The invalid node is removed when it is located at the starting point or the ending point, or when at least one of the preceding and following survey nodes is not a valid survey node.

[0008] Furthermore, data on ginseng black spot disease index, ginseng gray mold disease index, meteorological factor data, and preset early warning thresholds are obtained, including: Acquire the meteorological factor data and perform quality control on the meteorological factor data; The meteorological factor data includes daily meteorological factor data, and the quality control includes: Outlier nodes that exceed the preset physical extreme value boundary and null value nodes caused by disconnection in the daily meteorological factor data are marked as invalid records, and observations that do not exceed the preset physical extreme value boundary and are not null values ​​are determined as valid observations. The preset physical extreme boundary includes an atmospheric temperature of -20℃ to 50℃, a relative humidity of 0% to 100%, and rainfall and sunshine hours of not less than 0. When the number of consecutive missing days of the invalid record is less than the preset meteorological factor data missing tolerance threshold, linear interpolation is performed using the valid observation values ​​of adjacent dates. When the number of consecutive missing days is greater than or equal to the meteorological factor data missing tolerance threshold, an alarm is triggered and the invalid record is isolated. The daily meteorological factor data after quality control is converted into meteorological factor data for each ten-day period.

[0009] Furthermore, the disease index for mid-August is determined from the ginseng black spot disease disease index data and the ginseng gray mold disease disease index data, including: Filter the disease index records whose survey date is in August, whose survey period is in the middle of August, and whose survey node is a valid survey node to form valid disease index records; When there are multiple valid disease index records with the same disease name, location identifier, and year, the valid disease index record with the survey date closest to August 15th shall be selected. If there are two valid disease index records with the same number of days between them and August 15th, the arithmetic mean of the two valid disease index records is taken to determine the disease index of ginseng black spot disease in mid-August and the disease index of ginseng gray mold disease in mid-August, respectively.

[0010] Furthermore, the mid-August disease index includes the mid-August disease index of ginseng black spot disease and the mid-August disease index of ginseng gray mold disease, and the key meteorological factors include the key meteorological factors of ginseng black spot disease and the key meteorological factors of ginseng gray mold disease. The meteorological factor data involved in the calculation of the Pearson correlation coefficient include atmospheric temperature, relative humidity, rainfall and sunshine hours for each ten-day period from mid-May to late June, rainfall in mid-May and late June; The original training samples include ginseng black spot disease training samples and ginseng gray mold disease training samples. The ginseng black spot disease training samples include key meteorological factors of ginseng black spot disease and the ginseng black spot disease disease index in mid-August, which have the same location identification and the same year. The ginseng gray mold disease training samples include key meteorological factors of ginseng gray mold disease and the ginseng gray mold disease disease index in mid-August, which have the same location identification and the same year.

[0011] Furthermore, generating enhanced training samples by superimposing random noise onto the original training samples includes: Obtain the preset number of augmented samples, feature value perturbation amplitude, and target value perturbation range; Random noise with an absolute value not greater than the perturbation amplitude of the feature value is superimposed on the feature value of each original training sample to generate enhanced feature values; Random noise within the perturbation range of the target value is superimposed on the target value of each original training sample to generate an enhanced target value; For each original training sample, an augmented training sample is generated according to the number of augmented samples, and the original training sample and the augmented training sample are merged to form the expanded training set.

[0012] Furthermore, the expanded training set is divided into a training part and a validation part, including: The expanded training sets include the ginseng black spot disease expanded training set and the ginseng gray mold disease expanded training set; The ginseng black spot disease extended training set is divided into a ginseng black spot disease training part and a ginseng black spot disease verification part, and the ginseng gray mold disease extended training set is divided into a ginseng gray mold disease training part and a ginseng gray mold disease verification part. Each of the original training samples and the enhanced training samples generated based on the original training samples are jointly assigned to the training part or jointly assigned to the verification part.

[0013] Furthermore, the candidate prediction models include gradient boosting regression, random forest regression, support vector regression, K-nearest neighbor regression, and ridge regression multinomial regression.

[0014] Furthermore, the preset warning threshold includes a low-risk upper limit and a medium-risk upper limit, wherein the low-risk upper limit is smaller than the medium-risk upper limit; Generate warning information based on the preset warning threshold, including: A low-risk warning is generated when the predicted value of the disease index does not exceed the low-risk upper limit; a medium-risk warning is generated when the predicted value of the disease index is greater than the low-risk upper limit but does not exceed the medium-risk upper limit; and a high-risk warning is generated when the predicted value of the disease index is greater than the medium-risk upper limit.

[0015] This invention provides a method for predicting and warning the risk of ginseng gray mold and black spot diseases based on meteorological factors. It has the following beneficial effects: 1. This invention utilizes a Python-developed automatic screening program for disease growth models, enabling rapid fitting and performance evaluation of various mathematical models. Compared to traditional manual fitting methods, this mechanism improves research efficiency and data processing accuracy. It identifies the cubic function model as the optimal model for fitting the temporal changes in the disease indices of ginseng black spot and gray mold, thus accurately revealing the natural development patterns of these two diseases over the course of the investigation.

[0016] 2. This invention uses the Pearson correlation coefficient method to screen key meteorological factors, clarifying the core meteorological driving factors affecting the occurrence of two diseases, and constructs a disease prediction model based on this. An automatic optimization mechanism determines gradient boosting regression as the optimal algorithm. Simultaneously, by combining data augmentation, control feature expansion, and regularization constraints, the problem of model overfitting under small field sample conditions is alleviated, improving the prediction accuracy and robustness of the disease occurrence severity in mid-August.

[0017] 3. This invention develops a prediction and early warning system using a front-end and back-end separation architecture. It embeds disease growth models and disease prediction models into the system, enabling unified management, querying, and visualization of meteorological data and disease index data. Based on meteorological factor data of the area to be predicted, the system automatically extracts features and outputs disease index prediction results and early warning information. With a user-friendly interface, it provides a reliable intelligent decision-making tool for early monitoring of ginseng diseases, analysis of epidemic trends, and the scientific formulation of application plans. Attached Figure Description

[0018] Figure 1This is the main flowchart of the method of the present invention; Figure 2 This is a system structure block diagram of the present invention; Figure 3 This is a flowchart of the field survey and data collection and processing of the present invention; Figure 4 This is a flowchart of the disease growth model fitting calculation of the present invention; Figure 5 This is a flowchart illustrating the key meteorological factor screening and model input feature construction process of this invention. Figure 6 This is a flowchart of the training data construction and data augmentation process of the present invention; Figure 7 This is a flowchart illustrating the feature standardization, prediction model training, and prediction warning output of the present invention. Figure 8 This is a bar chart comparing the actual severity index of ginseng black spot disease with the predicted value, as presented in this invention. Detailed Implementation

[0019] The technical solutions in 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.

[0020] See attached document Figure 1 This invention provides a method for predicting and warning the risk of ginseng gray mold and black spot based on meteorological factors, comprising the following steps: S10: The system receives survey leaf data collected every 3 to 7 days during the ginseng growing season using a five-point field sampling method. Each sampling point includes 5 ginseng plants, and the survey leaf data is evaluated using the survey leaf as the unit. Based on the survey leaf data, the system classifies the survey leaves for ginseng black spot disease and ginseng gray mold disease according to the proportion of diseased area to leaf area. The system calculates the ginseng black spot disease index and ginseng gray mold disease index based on the classification results, and receives data on four meteorological factors at the observation points: atmospheric temperature, relative humidity, rainfall, and sunshine duration. S20 uses the survey time series value as the independent variable and the arithmetic mean of the effective disease index of each observation point under the same survey time series value as the dependent variable. Curve fitting is performed on linear function, quadratic function, cubic function, exponential function, logarithmic function, power function, logistic function, Gompertz function and single-molecule model. The optimal disease growth model is selected based on the coefficient of determination, F statistic and standard deviation. The optimal disease growth model for ginseng black spot disease and ginseng gray mold disease is a cubic function model. The disease growth model is used to characterize the cumulative growth process of the disease index as the survey time series value changes. S30 uses the Pearson correlation coefficient method to calculate the correlation between the mid-August disease index of ginseng black spot disease and the mid-August disease index of ginseng gray mold disease and meteorological factors from mid-May to late June. The meteorological factors include atmospheric temperature, relative humidity, rainfall and sunshine hours. The system selects key meteorological factors from high to low based on the absolute value of the Pearson correlation coefficient and forms key meteorological factor sets for ginseng black spot disease and ginseng gray mold disease, respectively. S40: Using the key meteorological factors selected in step S30 as feature values ​​and the mid-August disease index as the target value, initial datasets for ginseng black spot disease and ginseng gray mold disease are constructed respectively. The system performs data augmentation on the initial datasets and standardizes the feature values ​​before model training. In the ridge regression multinomial regression branch, the system retains only the first-order terms and pairwise interaction terms, and adopts L2 regularization constraints to reduce the risk of overfitting under small sample training conditions. S50 uses gradient boosting regression, random forest regression, support vector regression, K-nearest neighbor regression, and ridge regression multinomial regression as candidate prediction models for training, and selects the candidate prediction model with the highest determination coefficient in the validation part as the optimal prediction model. The system embeds the disease growth model and disease prediction model into the early warning system, outputting the disease index prediction results and early warning information for ginseng black spot and ginseng gray mold.

[0021] See attached document Figure 2 The present invention also provides a risk prediction and early warning system for ginseng gray mold and black spot based on meteorological factors. The system includes a data management module, a disease growth model module, a disease prediction and early warning module, and a user authentication management module.

[0022] The data management module includes a meteorological factor data submodule and a disease index data submodule. The meteorological factor data submodule is used to store, query, add, modify, and delete four types of meteorological factor data: atmospheric temperature, relative humidity, rainfall, and sunshine duration. The disease index data submodule is used to store, query, add, modify, and delete ginseng black spot disease index data and ginseng gray mold disease index data. The data management module also includes a region management function, which is used to maintain the region, location identifier, and corresponding meteorological factor data source of the observation point, enabling meteorological factor data and disease index data to be associated according to the same location identifier.

[0023] The disease growth model module is used to call disease index time series data, fit linear functions, quadratic functions, cubic functions, exponential functions, logarithmic functions, power functions, logistic functions, Gompertz functions and single-molecule models, select the optimal disease growth model based on the coefficient of determination, F statistic and standard deviation, and output the fitted curve, fitted equation and fitted evaluation results.

[0024] The disease prediction and early warning module includes two sub-modules: one for predicting ginseng black spot disease and the other for predicting ginseng gray mold. These sub-modules extract key meteorological factors for ginseng black spot disease and ginseng gray mold disease, call the optimal prediction models for these two diseases, and output the predicted disease index value for mid-August, the name of the optimal prediction model used, the coefficient of determination of the optimal prediction model in the validation section, and early warning information. Specifically, the ginseng black spot disease prediction sub-module extracts key meteorological factors A1 to A5 for ginseng black spot disease, and the ginseng gray mold disease prediction sub-module extracts key meteorological factors B1 to B5 for ginseng gray mold disease. The meanings of A1 to A5 are determined in step S303, and the meanings of B1 to B5 are determined in step S304.

[0025] The user authentication management module is used to perform user login verification, user logout, user identity verification, and basic user information management. Users who fail to pass the login verification cannot access the data management module, the disease growth model module, and the disease prediction and early warning module.

[0026] The system adopts a front-end and back-end separation architecture. The back-end uses the Django framework to encapsulate database access and model calculation logic. Curve fitting calculations for disease growth models are implemented using the Python scipy library, while training and invocation of disease prediction models are implemented using the Python sklearn framework. A RESTful API interface is used to provide JSON format data services to the front-end. The front-end uses the Vue 3 framework for page rendering and interactive control. It sends requests to the back-end RESTful API interface and receives returned data via the Axios HTTP client. The user interface is built using the Element Plus component library, and the ECharts chart library is used to display disease indices, fitted curves, prediction results, and model comparison results. The database uses either MySQL or SQLite3.

[0027] The database is configured with a meteorological factor data information table and a disease index data information table. The meteorological factor data information table stores the year, month, location identifier, recording period, atmospheric temperature, relative humidity, rainfall, and sunshine hours. The location identifier is a unique identifier corresponding to the observation point. The recording period includes the first ten days of the month, the middle ten days, the last ten days, or the month. The meteorological factor data information table uses the year, month, location identifier, and recording period as a joint unique constraint, ensuring that the same location identifier, the same year, the same month, and the same recording period correspond to only one meteorological factor data record. The disease index data information table stores the disease name, location identifier, year, survey date, survey period, survey time series value, and disease index. The survey period is used to identify whether the survey date corresponds to the first ten days of the month, the middle ten days, or the last ten days. The system extracts the disease index for mid-August based on the disease name, location identifier, year, survey date, and survey period.

[0028] The mid-August disease index refers to valid disease index records where the survey date falls within August and the survey period is in the middle of the month. When multiple valid disease index records exist in mid-August for the same disease name, location, and year, the system selects the record with the survey date closest to August 15th as the mid-August disease index. When two valid disease index records exist with the same number of days between their survey dates and August 15th, the system takes the arithmetic mean of these two records as the mid-August disease index. The system uses the determined mid-August disease index as the target value for the corresponding disease prediction model.

[0029] The technical solution of the present invention will be described in detail below with reference to various detailed embodiments.

[0030] See attached document Figure 3 The method provided by the present invention includes the following sub-steps when performing step S10: S101, field investigators set up observation points in the ginseng planting area and conducted regular surveys using the five-point field sampling method during the ginseng growing season. The system received the leaf data generated from these regular surveys. In this embodiment, the survey frequency was once every 3 to 7 days. The five-point field sampling method involved selecting the center point and four corner areas within the target observation area as sampling points, with 5 ginseng plants selected at each sampling point as survey samples. The survey used the leaf as the evaluation unit, recording the disease severity of ginseng black spot and ginseng gray mold, respectively.

[0031] S102, the surveyed leaves are graded according to the proportion of lesion area to leaf area. The same grading standard is used for ginseng black spot and ginseng gray mold: Grade 0 is recorded when the surveyed leaves have no lesions; Grade 1 is recorded when the proportion of lesions to leaf area is greater than 0 but not more than 1.0%; Grade 3 is recorded when the proportion of lesions to leaf area is 1.1% to 10.0%; Grade 5 is recorded when the proportion of lesions to leaf area is 10.1% to 25.0%; Grade 7 is recorded when the proportion of lesions to leaf area is 25.1% to 50.0%; and Grade 9 is recorded when the proportion of lesions to leaf area is greater than 50.0%. The system counts the number of surveyed leaves at each grade based on this grading standard.

[0032] S103, calculate the disease index for ginseng black spot and ginseng gray mold based on the number of leaves surveyed at each level. The disease index is used to characterize the severity of disease occurrence within the same survey batch, and the calculation formula is: ; In the formula, This refers to the disease severity index; This represents the total number of disease severity levels, with a value of 6. Index for disease severity grading; For the first The number of surveyed leaves corresponding to each disease severity level; For the first The disease level number corresponding to each disease severity level is set to {0, 1, 3, 5, 7, 9}. To investigate the total number of leaves; This represents the highest severity level, with a value of 9. When... When the value is 0 or missing, the system will mark the corresponding survey node as an invalid node and it will not directly participate in the calculation of the disease index; when When the value is greater than 0, the system will mark the corresponding survey node as a valid survey node.

[0033] S104 synchronously collects meteorological factor data for atmospheric temperature, relative humidity, rainfall, and sunshine duration at the observation points. The meteorological factor data originates from observation records at meteorological stations near each survey point, and the meteorological factor data is correlated with disease index data according to the survey period. After receiving the daily meteorological factor data, the system converts it into ten-day meteorological factor data. Atmospheric temperature and relative humidity are averaged based on valid observations within the ten-day period, while rainfall and sunshine duration are summed based on valid observations within the ten-day period. The converted ten-day meteorological factor data serves as input data for subsequent key meteorological factor screening and disease prediction model training. When the key meteorological factor is rainfall spanning ten days or monthly rainfall, the system obtains it by summing the daily rainfall data within the corresponding time range; when the system input data is already ten-day rainfall, the system sums the corresponding ten-day rainfall to obtain the ten-day or monthly rainfall.

[0034] The method provided by this invention, when grading sampled survey leaves according to the proportion of lesion area to leaf area, specifically includes the following sub-steps: S105, Obtain the whole leaf area and lesion area of ​​a single surveyed leaf. For ginseng plant leaves obtained from field sampling, area data is extracted as the input basis for grading and evaluation. For obtaining the whole leaf area and lesion area of ​​a single surveyed leaf, those skilled in the art can use a transparent grid comparison method to obtain a physically estimated area, or use an image acquisition device to capture leaf images and run an image segmentation algorithm to calculate the lesion pixel area and whole leaf pixel area. These area quantification methods are well-known techniques in the field and will not be elaborated upon here.

[0035] S106, calculate the percentage of leaf area covered by lesions. After receiving the whole leaf area data and the lesion area data, the system divides the lesion area by the whole leaf area to obtain a percentage value representing the severity of the disease.

[0036] S107: The percentage of lesion area is compared and mapped with the disease grading standard in step S102 to generate the disease grade corresponding to the currently investigated leaf. To avoid inconsistencies in interval boundary judgments, before performing the comparison and mapping, the system uniformly retains the percentage of lesion area calculated in step S106 to one decimal place and determines a unique disease grade according to the disease grading standard in step S102. When the percentage of lesion area is 0, the system classifies the corresponding investigated leaf as grade 0; when the percentage of lesion area is greater than 0 but does not exceed 1.0%, the system classifies the corresponding investigated leaf as grade 1.

[0037] S108: Based on the mapping comparison results, the surveyed leaves at each level are categorized and counted. According to the judgment result of step S107, the system generates a corresponding disease level label for the currently assessed surveyed leaves. Then, it iterates through all sample labels within the current sampling batch at the current observation point, counts the total number of surveyed leaves belonging to each disease level, and generates a set of surveyed leaf counts for each level. This statistical data is pushed to the system's data management module for storage, serving as data input for the quantitative calculation of the field disease index.

[0038] The method provided by this invention, when calculating the field disease index based on grading records, specifically includes the following sub-steps: S109, Obtain disease severity grading statistics and grading parameters. In a specific implementation scenario, the calculation and processing unit reads the set of leaf counts for each level of the survey generated by the disease severity index data submodule of the data management module, and simultaneously summarizes the total number of leaves sampled in the current batch. At the same time, the calculation and processing unit obtains preset disease severity grading parameters, which include the total number of disease severity grading levels and the corresponding disease level sequence.

[0039] S110 performs a weighted cumulative calculation of disease severity. The calculation unit multiplies the number of surveyed leaves for each disease severity level with its corresponding disease level, and then sums the products of all disease severity levels to obtain a weighted value reflecting the overall severity of ginseng black spot and ginseng gray mold at the current observation point. The higher the disease level, the greater the weight of the corresponding surveyed leaves in the disease index calculation, thus characterizing the degree of disease occurrence within the current survey batch.

[0040] S111, Generate the disease severity index based on normalization logic. The system converts the weighted cumulative result obtained in step S110 into a percentage-based disease severity index according to the disease severity index calculation formula in step S103. To ensure consistency in the processing logic before and after, when investigating the total number of leaves... When the value is 0 or missing, the system first marks the survey node as an invalid node; this invalid node does not participate in the calculation of the original disease index. Before fitting the disease growth model, if the invalid node is located in the middle of the disease index time series, and the previous and next survey nodes of the invalid node are both valid survey nodes, the system uses linear interpolation to complete the invalid node; if the invalid node is located at the beginning or end of the disease index time series, or if at least one of the previous and next survey nodes of the invalid node is not a valid survey node, the system removes the invalid node.

[0041] S112, Generate a time series of disease indexes. The computational processing unit correlates the calculated disease index values ​​for ginseng black spot and ginseng gray mold with the physical timestamp of the current field survey, generating disease index records that include a time dimension. The generated records are transmitted to the data management module for storage. Within the set growing season cycle, the disease index records acquired continuously at time intervals together constitute a time series characterizing the natural development dynamics of diseases in the field, serving as direct data input for triggering subsequent multi-time series mathematical model fitting operations.

[0042] The method provided by this invention, when performing synchronous acquisition of meteorological factor data from meteorological stations at various observation points, specifically includes the following sub-steps: S113, Acquire basic meteorological factor data. The system synchronously receives meteorological factor data from the field environment through meteorological observation stations deployed near ginseng planting observation points. In the embodiments of this invention, the collected data dimensions mainly include four items: atmospheric temperature, relative humidity, rainfall, and sunshine duration. For the acquisition and wireless transmission of meteorological factor data, those skilled in the art can use conventional temperature and humidity sensors, tipping bucket rain gauges, and photoelectric sunshine meters combined with data transmission modules. The physical sensing and signal conversion mechanisms are well-known technologies in the field and will not be described in detail here.

[0043] S114, Perform quality control on meteorological factor time-series data. In some implementations, the computing unit initiates a threshold verification procedure after receiving the raw daily meteorological factor data. The system sets preset physical extreme value boundaries for each meteorological factor based on the natural climate patterns of the local ginseng growing season. In this embodiment, the preset physical extreme value boundaries include an effective range of atmospheric temperature of -20℃ to 50℃, an effective range of relative humidity of 0% to 100%, and a lower limit of 0 for rainfall and sunshine hours. When outliers exceeding the preset physical extreme value boundaries or null nodes caused by disconnection are detected, the system marks the outliers or null nodes as invalid records. Subsequently, the system determines the number of consecutive missing days for invalid records. If the number of consecutive missing days is less than the meteorological factor data missing tolerance threshold stored in the system parameter table, linear interpolation is performed using the effective observation values ​​of adjacent dates; if the number of consecutive missing days is greater than or equal to the meteorological factor data missing tolerance threshold, the system triggers an alarm and isolates invalid records, which are not included in the key meteorological factor screening and disease prediction model training process.

[0044] S115 generates meteorological factor data for each ten-day period. The records returned by the underlying sensors are typically daily meteorological factor data. However, the development of ginseng black spot and ginseng gray mold exhibits a long-term cumulative effect with meteorological conditions. At the operational level, down-converting and aggregating the high-frequency fluctuating daily meteorological factor data can filter out noise caused by short-term meteorological disturbances and extract long-term environmental characteristics that match the incubation and infection cycles of pathogens. Therefore, the system performs time-window aggregation processing on the quality-controlled daily meteorological factor data.

[0045] The system divides a calendar month into three periods: the first ten days, the middle ten days, and the last ten days. The specific calendar division rules are as follows: the 1st to the 10th of each month is the first ten days, the 11th to the 20th is the middle ten days, and the 21st to the last day of the month is the last ten days. For state-type meteorological factors such as atmospheric temperature and relative humidity, the calculation and processing unit averages the daily effective observations within the ten-day period to obtain the ten-day average atmospheric temperature and relative humidity. For cumulative meteorological factors such as rainfall and sunshine hours, the calculation and processing unit sums the daily effective observations within the ten-day period to obtain the ten-day rainfall and sunshine hours.

[0046] The system uses the above aggregation operations to convert daily meteorological factor data into sets of meteorological factor data for each ten-day period, and saves them in the data management module as the basic feature input for subsequent calculation of the Pearson correlation coefficient. Rainfall spanning ten days and monthly rainfall are obtained by summing the daily rainfall data within the corresponding time range; when the system input data is already ten-day rainfall, the rainfall spanning ten days and monthly rainfall are obtained by summing the rainfall within the corresponding ten-day period.

[0047] See attached document Figure 4 The method provided by this invention, when performing fitting calculations of disease indexes using multiple time-series mathematical models, specifically includes the following sub-steps: S201, Extract the time series of disease index and perform time axis alignment. The system reads the time series data of ginseng black spot disease index and ginseng gray mold disease index from the data management module, sets the first survey node as the time origin, records the survey time series value of the first survey node as 0, and records the number of days between subsequent survey nodes and the first survey node as the survey time series value. In this invention, This only represents the time series values ​​of the survey, in days, and It is used only as an independent variable in the disease growth model, and not as an input feature in the disease prediction model.

[0048] S202, Construct a set of alternative growth models. The system uses the survey time series value t as the independent variable and the arithmetic mean of the effective disease index of each observation point under the same survey time series value as the dependent variable, and performs curve fitting on linear function, quadratic function, cubic function, exponential function, logarithmic function, power function, logistic function, Gompertz function and single-molecule model respectively.

[0049] S203, Perform growth model fitting and evaluation. The system calculates the coefficient of determination, F-statistic, and standard deviation of each candidate growth model, and uses the coefficient of determination as the primary screening indicator. When the coefficients of determination of two candidate growth models are the same, the system further compares the standard deviations, and the one with the smaller standard deviation is given priority. When both the coefficient of determination and the standard deviation are the same, the system compares the F-statistics, and the one with the larger F-statistic is given priority.

[0050] S204, Determine the optimal disease growth model. Through fitting comparison, the optimal disease growth models for both ginseng black spot and ginseng gray mold are cubic function models.

[0051] The optimal disease growth model for ginseng black spot disease is: ; The optimal disease growth model for ginseng gray mold is: ; In the formula, The severity index for ginseng black spot disease. As a disease index for ginseng gray mold, These are the time series values ​​for the survey.

[0052] S205. Disease occurrence dynamics are determined based on the optimal disease growth model and field survey data. Occurrence dynamics are used to interpret the fitting results of the disease growth model. The disease prediction model still uses the disease index in mid-August as the target value. The disease growth model does not replace the disease prediction model.

[0053] S206, Save the disease growth model results. The system stores the optimal disease growth model name, model expression, coefficient of determination, F-statistic, and standard deviation in the data management module, and outputs the fitted curve and fitting evaluation results to the disease growth model module. The disease growth model is used to describe the cumulative growth process of the disease index as a function of the survey time series values; its input variable is the survey time series value. The disease prediction model is used to predict the disease index in mid-August based on key meteorological factors. Its input variables are key meteorological factors A1 to A5 for ginseng black spot disease or key meteorological factors B1 to B5 for ginseng gray mold disease. The disease growth model does not replace the disease prediction model in steps S40 and S50.

[0054] See attached document Figure 5The method provided by this invention, when performing key meteorological factor screening and model input feature construction, specifically includes the following sub-steps: S301, Determine the key meteorological factors for screening. The system identifies the mid-August disease index as the target value and meteorological factors from mid-May to late June as candidate feature values. Meteorological factors include atmospheric temperature, relative humidity, rainfall, and sunshine duration. Atmospheric temperature and relative humidity are averaged based on valid observations within the corresponding ten-day period, while rainfall and sunshine duration are summed based on valid observations within the corresponding ten-day period. When candidate feature values ​​involve rainfall spanning multiple ten-day periods or monthly rainfall, the system generates corresponding feature values ​​according to the aggregation criteria determined in step S104.

[0055] S302, Calculate the Pearson correlation coefficient. The system calculates the Pearson correlation coefficients between the mid-August disease index of ginseng black spot disease and meteorological factors, cross-ten-day rainfall, and monthly rainfall from mid-May to late June, as well as the mid-August disease index of ginseng gray mold disease and meteorological factors, cross-ten-day rainfall, and monthly rainfall from mid-May to late June. The meteorological factors are sorted from highest to lowest absolute value according to the Pearson correlation coefficient, and the top five meteorological factors are selected to form the key meteorological factors for ginseng black spot disease and ginseng gray mold disease.

[0056] S303, Screening Key Meteorological Factors for Ginseng Black Spot Disease. The key meteorological factors for ginseng black spot disease screened by the system include: rainfall in late May, rainfall in late June, average relative humidity in early June, sunshine hours in mid-May, and average relative humidity in late May. For ease of model input, the system records the rainfall in late May as A1, the rainfall in late June as A2, the average relative humidity in early June as A3, the sunshine hours in mid-May as A4, and the average relative humidity in late May as A5. In this invention, A1 to A5 are only used to represent the key meteorological factors for ginseng black spot disease and do not have any other meaning.

[0057] S304, Screening key meteorological factors for ginseng gray mold. The key meteorological factors for ginseng gray mold screened by the system include: rainfall in mid-to-late May, rainfall in late May, rainfall in late June, rainfall in June, and sunshine hours in mid-June. Among them, the rainfall in mid-to-late May and the rainfall in June are generated according to the aggregation caliber used in the ginseng gray mold training data construction phase; in this embodiment, the rainfall in mid-to-late May is the sum of the rainfall in mid-May and the rainfall in late May, and the rainfall in June is the sum of the rainfall in early June, the rainfall in mid-June, and the rainfall in late June. For ease of model input, the system records the rainfall in mid-to-late May as B1, the rainfall in late May as B2, the rainfall in late June as B3, the rainfall in June as B4, and the sunshine hours in mid-June as B5. In this invention, B1 to B5 are only used to represent key meteorological factors for ginseng gray mold and do not have any other meaning.

[0058] S305, Generate a set of key meteorological factors. The system maps A1 to A5 to the mid-August disease index of ginseng black spot disease to form ginseng black spot disease training data, and maps B1 to B5 to the mid-August disease index of ginseng gray mold disease to form ginseng gray mold disease training data. Each ginseng black spot disease training sample corresponds to the same location identifier and the same year for the key meteorological factors A1 to A5 of ginseng black spot disease and their corresponding mid-August disease index; each ginseng gray mold disease training sample corresponds to the same location identifier and the same year for the key meteorological factors B1 to B5 of ginseng gray mold disease and their corresponding mid-August disease index. The system does not perform principal component dimensionality reduction on the above key meteorological factors to ensure that the input feature names are consistent with the Pearson correlation coefficient screening results. When there is a temporal range inclusion relationship among the key meteorological factors B1, B2, and B4 for ginseng gray mold, the system still retains B1, B2, and B4 according to the screening results of Pearson correlation coefficient; where B1 represents the rainfall in mid-to-late May, B2 represents the rainfall in late May, and B4 represents the rainfall in June, and the three correspond to the rainfall impact in different time ranges.

[0059] See attached document Figure 6 The method provided by this invention, when performing training data construction and data augmentation processing, specifically includes the following sub-steps: S401, Extract initial training data. The system reads the ginseng black spot disease training data and ginseng gray mold disease training data generated in step S305. The ginseng black spot disease training data uses A1 to A5 as feature values ​​and the mid-August ginseng black spot disease disease index as the target value; the ginseng gray mold disease training data uses B1 to B5 as feature values ​​and the mid-August ginseng gray mold disease disease index as the target value.

[0060] S402, Generate Enhanced Feature Values. The system reads the number of enhanced samples and the feature value perturbation amplitude stored in the system parameter table. Random noise is superimposed on each original key meteorological factor value, with the absolute value of the random noise not exceeding the feature value perturbation amplitude. Enhanced feature values ​​are generated to expand the training samples without changing the name and physical meaning of the original key meteorological factors. Enhanced feature values ​​must not exceed the physical value boundaries of the corresponding meteorological factor. The physical value boundaries include an atmospheric temperature of -20℃ to 50℃, a relative humidity of 0% to 100%, and rainfall and sunshine duration not less than 0. When an enhanced feature value exceeds the physical value boundaries of the corresponding meteorological factor, the system corrects it according to the physical value boundaries of that meteorological factor.

[0061] S403, Generate Enhanced Target Value. The system reads the target value perturbation range stored in the system parameter table, and superimposes random noise within the target value perturbation range onto the original disease index to generate the enhanced target value. The enhanced target value is limited to the range of 0 to 100; when the enhanced target value is less than 0, it is corrected to 0, and when the enhanced target value is greater than 100, it is corrected to 100.

[0062] S404, Forming an Expanded Training Set. The system merges the original training data and the enhanced training data to form an expanded training set. This expanded training set is used only for training and comparing disease prediction models; it does not change the fitting results of the disease growth model in step S20, nor does it change the selection results of key meteorological factors in step S30. The feature names in the original training data and the enhanced training data remain consistent, both using the key meteorological factor names determined in steps S303 and S304. The same original training sample and its corresponding enhanced training sample are assigned to the same training part or the same validation part to avoid samples from the same source entering both the training and validation parts simultaneously. During model evaluation, the system uses the validation part to calculate the coefficient of determination and compares candidate prediction models based on the coefficient of determination in the validation part.

[0063] See attached document Figure 7 The method provided by this invention, when performing feature standardization, prediction model training, and prediction warning output, specifically includes the following sub-steps: S501, Extract the expanded training set. The system reads the expanded training set formed in step S404 from the data management module and constructs the ginseng black spot disease feature matrix, ginseng black spot disease target value sequence, ginseng gray mold disease feature matrix, and ginseng gray mold disease target value sequence, respectively. The ginseng black spot disease feature matrix consists of A1 to A5, the ginseng gray mold disease feature matrix consists of B1 to B5, and the target value sequence is the disease index in mid-August. According to the preset sample partitioning rules stored in the system parameter table, the system divides the ginseng black spot disease expanded training set into a ginseng black spot disease training part and a ginseng black spot disease validation part, and the ginseng gray mold disease expanded training set into a ginseng gray mold disease training part and a ginseng gray mold disease validation part.

[0064] S502, Perform feature standardization processing. The system uses StandardScaler to fit standardized parameters to the training parts of ginseng black spot disease and ginseng gray mold, respectively. The corresponding standardized parameters are then used to process the key meteorological factor data of the training, validation, and prediction areas, ensuring a uniform data scale for each key meteorological factor before model training and prediction. Specifically, the standardized parameters for the ginseng black spot disease training stage are obtained only from the ginseng black spot disease training part and are used to process the key meteorological factors A1 to A5 of ginseng black spot disease in the training, validation, and prediction areas. Similarly, the standardized parameters for the ginseng gray mold disease training stage are obtained only from the ginseng gray mold disease training part and are used to process the key meteorological factors B1 to B5 of ginseng gray mold disease in the training, validation, and prediction areas.

[0065] The system saves the standardized parameters generated during the standardization process and uses these standardized parameters, saved during the corresponding disease training phase, to process key meteorological factor data for the area to be predicted during the disease prediction phase. For example, the ginseng black spot disease prediction phase only uses the standardized parameters saved during the ginseng black spot disease training phase, and the ginseng gray mold disease prediction phase only uses the standardized parameters saved during the ginseng gray mold disease training phase.

[0066] S503, Constructing Candidate Prediction Models. The system employs gradient boosting regression, random forest regression, support vector regression, K-nearest neighbor regression, and ridge regression multinomial regression as candidate prediction models. For ridge regression multinomial regression, the system sets the `interaction_only` parameter of `PolynomialFeatures` to true, ensuring that the feature expansion results retain only first-order terms and pairwise interaction terms, avoiding the generation of higher-order power terms for single features. The L2 regularization constraint of ridge regression is used to reduce the risk of overfitting. Decision tree regression can be used as an additional candidate prediction model in the model comparison, but the setting of gradient boosting regression, random forest regression, support vector regression, K-nearest neighbor regression, and ridge regression multinomial regression as the primary candidate prediction models remains unchanged.

[0067] S504, Train and evaluate candidate prediction models. The system trains and evaluates candidate prediction models based on the training and validation parts of the extended training sets for ginseng black spot and ginseng gray mold, respectively. In this embodiment, the system uses the ginseng black spot training part obtained in step S501 to train the ginseng black spot candidate prediction model and uses the ginseng black spot validation part to calculate the determination coefficient of the ginseng black spot candidate prediction model; the system uses the ginseng gray mold training part obtained in step S501 to train the ginseng gray mold candidate prediction model and uses the ginseng gray mold validation part to calculate the determination coefficient of the ginseng gray mold candidate prediction model. The system selects the candidate prediction model with the highest determination coefficient in the validation part for the corresponding disease as the optimal prediction model for the corresponding disease.

[0068] S505, determine the optimal prediction models for ginseng black spot disease and ginseng gray mold disease. By comparing models, the system identifies the best-performing models in predicting ginseng black spot disease and ginseng gray mold disease. The candidate prediction models with the highest determination coefficients in the validation phase are then determined as the optimal prediction models for ginseng black spot disease and ginseng gray mold disease, respectively.

[0069] S506, Output Prediction Results. After receiving key meteorological factor data for the region to be predicted from May to June, the system extracts key meteorological factors A1 to A5 for ginseng black spot disease and key meteorological factors B1 to B5 for ginseng gray mold disease. The system uses standardized parameters saved during the training phase of ginseng black spot disease to standardize A1 to A5, and inputs the standardized A1 to A5 into the optimal prediction model for ginseng black spot disease, outputting the predicted disease index value for ginseng black spot disease in mid-August. The system also uses standardized parameters saved during the training phase of ginseng gray mold disease to standardize B1 to B5, and inputs the standardized B1 to B5 into the optimal prediction model for ginseng gray mold disease, outputting the predicted disease index value for ginseng gray mold disease in mid-August. When the area to be predicted lacks meteorological factor data required for the formation of key meteorological factors A1 to A5 in ginseng black spot disease, the system outputs a message indicating insufficient prediction data for ginseng black spot disease and stops predicting the disease index of ginseng black spot disease in mid-August; when the area to be predicted lacks meteorological factor data required for the formation of key meteorological factors B1 to B5 in ginseng gray mold disease, the system outputs a message indicating insufficient prediction data for ginseng gray mold disease and stops predicting the disease index of ginseng gray mold disease in mid-August.

[0070] S507, Generate Early Warning Information. The system generates early warning information based on the predicted disease index and preset early warning thresholds. The early warning information includes the disease name, the predicted target period, the names and values ​​of key meteorological factors, the predicted disease index, the name of the optimal prediction model, and the early warning level. In this embodiment, the early warning thresholds are stored in the system parameter table. The early warning thresholds include a low-risk upper limit and a medium-risk upper limit, with the low-risk upper limit being less than the medium-risk upper limit. The low-risk and medium-risk upper limits are determined based on historical disease index data, regional production management requirements, or user configuration. When the predicted disease index does not exceed the low-risk upper limit, the system generates a low-risk early warning; when the predicted disease index exceeds the low-risk upper limit but does not exceed the medium-risk upper limit, the system generates a medium-risk early warning; when the predicted disease index exceeds the medium-risk upper limit, the system generates a high-risk early warning. When the system parameter table does not set a low-risk or medium-risk upper limit, the system only outputs the predicted disease index, the names and values ​​of key meteorological factors, the name of the optimal prediction model, and the coefficient of determination, without generating an early warning level.

[0071] S508, Save Prediction and Early Warning Records. The system saves the input key meteorological factor data, disease index prediction values, optimal prediction model name, coefficient of determination of the optimal prediction model in the validation part, and early warning information to the data management module, and visualizes the prediction results through the disease prediction and early warning module. The visualization includes the disease index prediction value, optimal prediction model name, coefficient of determination, early warning level, and the comparison results between the actual values ​​of the training samples and the model prediction values. When the system does not generate an early warning level, the visualization does not display the early warning level, or displays the early warning level as not configured.

[0072] Specific application examples: To further clarify the data processing flow logic of the prediction and early warning system of the present invention in a real production environment, the prediction processing flow of the present invention will be described using actual field monitoring and meteorological data from Fusong area, Jilin Province in 2024 as an example.

[0073] The system receives key meteorological factor data for the region to be predicted from May to June. Specifically, for ginseng black spot disease, the key meteorological factors extracted by the system are: late May rainfall (A1) 38.5 mm, late June rainfall (A2) 68.5 mm, early June average relative humidity (A3) 73.8%, mid-May sunshine hours (A4) 54.2 h, and late May average relative humidity (A5) 65.7%. Before prediction, the data scale for each factor needs to be standardized. The system calls the StandardScaler parameters saved during the training phase and, based on the expanded training set data for the region, calculates the historical mean of late May rainfall (A1) to be approximately 35.28 mm, with a standard deviation of approximately 3.61. Using the standardization formula, the standardized eigenvalue of A1 is approximately (38.5 - 35.28) / 3.61 ≈ 0.892. The remaining eigenfactors A2 to A5 are processed synchronously to form a standardized feature matrix.

[0074] The feature matrix was then input into the ginseng black spot disease prediction model. The optimal prediction model for ginseng black spot disease had previously undergone data augmentation by adding random noise and constrained the feature dimensions using PolynomialFeatures (interaction_only=True) and Ridge Regression L2 regularization. The gradient boosting regression algorithm was used for calculation, and the system output a predicted ginseng black spot disease severity index of 14.69 for the Fusong area in mid-August 2024. The system compared this predicted value with the preset low-risk and medium-risk upper limits in the parameter table, and generated a warning level after confirming that it fell within the corresponding range. Finally, the system front-end output, through the disease prediction and warning module, information including the names and values ​​of key meteorological factors, the predicted value of 14.69, the optimal prediction model name (gradient boosting regression), the model's coefficient of determination in the validation section, and a visualization of the corresponding warning level.

[0075] The following section uses experimental data to verify the predictive performance of the above model.

[0076] See attached document Figure 8This graph covers data from 2024 to 2025 from three observation points in Changchun, Fusong, and Dunhua. The horizontal axis represents the observation point and year, and the vertical axis represents the disease index. Each test sample in the graph contains two adjacent bars, representing the actual disease index in mid-August from the field survey and the model's predicted value, respectively. The close fit between the two bars shows that, with only six valid observation samples initially, the system expanded the training set using random noise data augmentation technology without significant overfitting. Quantitative data shows that the gradient boosting regression model predicted a disease index of 14.69 for the Fusong area in 2024, achieving an accuracy of 91.71%; the prediction accuracy for the Dunhua area in 2024 was 96.03%, and the average prediction accuracy for all observation points reached 93.24%. These experimental results demonstrate that the data augmentation and multi-model optimization mechanism used in this study can stably output predicted disease incidence results under limited field sample conditions.

Claims

1. A method for predicting and warning the risk of ginseng gray mold and black spot based on meteorological factors, characterized in that, include: Acquire ginseng black spot disease disease index data, ginseng gray mold disease disease index data, meteorological factor data, and preset early warning thresholds; The disease index for mid-August was determined from the ginseng black spot disease disease index data and the ginseng gray mold disease disease index data, respectively. Calculate the Pearson correlation coefficient between the mid-August disease index and the meteorological factor data respectively, and select the top five in absolute value from high to low to form the key meteorological factors; The key meteorological factors are used as feature values, and the mid-August disease index is used as the target value to form the original training sample. Random noise is superimposed on the original training samples to generate enhanced training samples. The original training samples and the enhanced training samples are merged to form an expanded training set. The expanded training set is divided into a training part and a validation part. Candidate prediction models are trained, and the candidate prediction models with the highest determination coefficients in the validation part are determined as the optimal prediction models. Receive meteorological factor data of the area to be predicted, extract corresponding feature data from the meteorological factor data according to the key meteorological factors and input them into the optimal prediction model to generate a disease index prediction value, and generate early warning information according to the preset early warning threshold.

2. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 1, characterized in that, Acquire ginseng black spot disease disease index data, ginseng gray mold disease disease index data, meteorological factor data, and preset early warning thresholds, including: Obtain the ginseng black spot disease disease index data and the ginseng gray mold disease disease index data; Receive survey leaf data generated using the five-point field sampling method; Obtain the whole leaf area and lesion area of ​​a single surveyed leaf from the surveyed leaf data, and determine the percentage value of lesion area and disease level; When the percentage values ​​of the lesion area are 0, greater than 0 and not exceeding 1.0%, 1.1% to 10.0%, 10.1% to 25.0%, 25.1% to 50.0%, and greater than 50.0%, the disease grades are 0, 1, 3, 5, 7, and 9, respectively. The number of leaves surveyed at each level and the total number of leaves surveyed were counted to generate time series data of ginseng black spot disease index and ginseng gray mold disease index, respectively. The time-series data of ginseng black spot disease index and the time-series data of ginseng gray mold disease index both include survey nodes consisting of disease name, location identifier, year, survey date, survey period, and disease index; Survey nodes with a total number of leaves of 0 or missing are marked as invalid nodes, and survey nodes with a total number of leaves greater than 0 are marked as valid survey nodes.

3. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 2, characterized in that, After generating time-series data on the ginseng black spot disease index and the ginseng gray mold disease index respectively, the following is also included: The survey nodes in the time series data of ginseng black spot disease index and the time series data of ginseng gray mold disease index are sorted according to the survey date, and the first and last survey nodes after sorting are determined as the start point and the end point, respectively. When the invalid node is neither located at the starting point nor at the ending point, and both the preceding and following survey nodes of the invalid node are valid survey nodes, the invalid node is completed using linear interpolation according to the survey dates of the preceding and following survey nodes. The invalid node is removed when it is located at the starting point or the ending point, or when at least one of the preceding and following survey nodes is not a valid survey node.

4. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 1, characterized in that, Acquire ginseng black spot disease disease index data, ginseng gray mold disease disease index data, meteorological factor data, and preset early warning thresholds, including: Acquire the meteorological factor data and perform quality control on the meteorological factor data; The meteorological factor data includes daily meteorological factor data, and the quality control includes: Outlier nodes that exceed the preset physical extreme value boundary and null value nodes caused by disconnection in the daily meteorological factor data are marked as invalid records, and observations that do not exceed the preset physical extreme value boundary and are not null values ​​are determined as valid observations. The preset physical extreme boundary includes an atmospheric temperature of -20℃ to 50℃, a relative humidity of 0% to 100%, and rainfall and sunshine hours of not less than 0. When the number of consecutive missing days of the invalid record is less than the preset meteorological factor data missing tolerance threshold, linear interpolation is performed using the valid observation values ​​of adjacent dates. When the number of consecutive missing days is greater than or equal to the meteorological factor data missing tolerance threshold, an alarm is triggered and the invalid record is isolated. The daily meteorological factor data after quality control is converted into meteorological factor data for each ten-day period.

5. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 2, characterized in that, The disease index for mid-August was determined from the ginseng black spot disease disease index data and the ginseng gray mold disease disease index data, including: The disease index records that are surveyed in August, during the middle of August, and at the valid survey node are selected to form valid disease index records. When there are multiple valid disease index records with the same disease name, location identifier, and year, the valid disease index record with the survey date closest to August 15th shall be selected. If there are two valid disease index records with the same number of days between them and August 15th, the arithmetic mean of the two valid disease index records is taken to determine the disease index of ginseng black spot disease in mid-August and the disease index of ginseng gray mold disease in mid-August, respectively.

6. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 5, characterized in that, The mid-August disease index includes the mid-August disease index of ginseng black spot disease and the mid-August disease index of ginseng gray mold disease; the key meteorological factors include the key meteorological factors of ginseng black spot disease and the key meteorological factors of ginseng gray mold disease. The meteorological factor data involved in the calculation of the Pearson correlation coefficient include atmospheric temperature, relative humidity, rainfall and sunshine hours for each ten-day period from mid-May to late June, rainfall in mid-May and late June; The original training samples include ginseng black spot disease training samples and ginseng gray mold disease training samples. The ginseng black spot disease training samples include key meteorological factors of ginseng black spot disease and the ginseng black spot disease disease index in mid-August, which have the same location identification and the same year. The ginseng gray mold disease training samples include key meteorological factors of ginseng gray mold disease and the ginseng gray mold disease disease index in mid-August, which have the same location identification and the same year.

7. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 6, characterized in that, Adding random noise to the original training samples to generate enhanced training samples includes: Obtain the preset number of augmented samples, feature value perturbation amplitude, and target value perturbation range; Random noise with an absolute value not greater than the perturbation amplitude of the feature value is superimposed on the feature value of each original training sample to generate enhanced feature values; Random noise within the perturbation range of the target value is superimposed on the target value of each original training sample to generate an enhanced target value; For each original training sample, an augmented training sample is generated according to the number of augmented samples, and the original training sample and the augmented training sample are merged to form the expanded training set.

8. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 7, characterized in that, The expanded training set is divided into a training part and a validation part, including: The expanded training sets include the ginseng black spot disease expanded training set and the ginseng gray mold disease expanded training set; The ginseng black spot disease extended training set is divided into a ginseng black spot disease training part and a ginseng black spot disease verification part, and the ginseng gray mold disease extended training set is divided into a ginseng gray mold disease training part and a ginseng gray mold disease verification part. Each of the original training samples and the enhanced training samples generated based on the original training samples are jointly assigned to the training part or jointly assigned to the verification part.

9. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 8, characterized in that, The candidate prediction models include gradient boosting regression, random forest regression, support vector regression, K-nearest neighbor regression, and ridge regression multinomial regression.

10. The method for predicting and warning the risk of ginseng gray mold and black spot disease based on meteorological factors according to claim 1, characterized in that, The preset warning threshold includes a low-risk upper limit and a medium-risk upper limit, wherein the low-risk upper limit is smaller than the medium-risk upper limit; Generate warning information based on the preset warning threshold, including: A low-risk warning is generated when the predicted value of the disease index does not exceed the low-risk upper limit; a medium-risk warning is generated when the predicted value of the disease index is greater than the low-risk upper limit but does not exceed the medium-risk upper limit; and a high-risk warning is generated when the predicted value of the disease index is greater than the medium-risk upper limit.

Images