Method for mapping large area of rice bacterial leaf blight severity by multi-source data collaboration of starlight earth
By combining drone and satellite imagery, a severity index for bacterial blight of rice was constructed. Through multi-model optimization, the problem of large-scale and accurate monitoring of bacterial blight of rice was solved, achieving high-precision disease inversion and mapping, and providing a scientific basis for monitoring.
Patent Information
- Application Number
- CN202510477922.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2025-12-26
- Estimated Expiration
- 2045-04-16
AI Technical Summary
Existing technologies make it difficult to achieve large-scale and accurate monitoring of rice bacterial blight. Satellite data has limited spatial/spectral resolution and low disease identification accuracy, while ground surveys are time-consuming and labor-intensive, making it difficult to obtain accurate disease indices.
By combining UAV multispectral imagery and Sentinel-2 imagery, and through multi-model comparison and cross-validation, a bacterial blight damage severity index (BLB_DSI) for rice was constructed. By integrating support vector machine (SVM), random forest (RF), and multiple linear regression (MLR) models, the severity of the disease can be accurately inverted and spatially mapped.
It has achieved high-precision, large-scale quantitative inversion and mapping of rice bacterial blight, which can accurately characterize the actual damage of the disease at the satellite pixel scale, providing scientific monitoring basis and technical support, and expanding the potential to other crop disease monitoring.
Smart Images
Figure CN120375237B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of remote sensing images, and more particularly to a method for mapping the severity of rice bacterial leaf blight on a large scale by using multi-source data from space, air and ground. BACKGROUND
[0002] Bacterial leaf blight (BLB) is one of the three major diseases affecting rice production. Traditional rice bacterial leaf blight monitoring mainly relies on manual field investigation, which not only consumes a large amount of manpower and time, but also is difficult to achieve large-scale monitoring, and cannot obtain the occurrence range and severity information of the disease in a timely and accurate manner.
[0003] With the development of remote sensing technology, it has been widely applied in the field of crop disease monitoring. Early research mainly focused on obtaining spectral data of crop leaves and canopies using ground observation equipment, and then using spectral indices, variance analysis, convolutional neural networks and other methods for disease feature extraction and classification. Although these methods have achieved high accuracy at the leaf / canopy scale, due to the limitations of ground data acquisition, it is difficult to achieve large-area disease monitoring. Unmanned aerial vehicles (UAVs) can obtain high spatial and spectral resolution remote sensing data and have gradually been applied in disease monitoring and achieved good results, but the monitoring range of UAVs is limited and it is difficult to meet the demand of large-scale monitoring.
[0004] Satellite data has the advantage of wide coverage and has become an important source of data for large-scale disease monitoring. However, the spatial / spectral resolution of satellite data is limited, and there is a problem of mixed pixels, resulting in relatively low disease identification accuracy. In addition, when using remote sensing data to quantify disease severity, it is usually necessary to obtain reference disease severity data, such as disease index (DI), but it is difficult to obtain accurate DI values, and the temporal and spatial resolution of existing satellite data also limits the accurate estimation of disease severity. Therefore, how to use multi-scale remote sensing data to accurately monitor the severity of rice bacterial leaf blight has become a problem to be solved in the field of agricultural disease and pest monitoring. SUMMARY
[0005] The present application aims to overcome the shortcomings of the prior art and provides a method for mapping the severity of rice bacterial leaf blight on a large scale by using multi-source data from space, air and ground.
[0006] In a first aspect, a method for mapping the severity of rice bacterial leaf blight on a large scale by using multi-source data from space, air and ground is provided, comprising:
[0007] S1, obtaining unmanned aerial vehicle multispectral images and Sentinel-2 images and preprocessing them;
[0008] S2, combine ground investigation data and visual interpretation to determine healthy and BLB damaged rice sample points in the unmanned aerial vehicle multispectral image; calculate a plurality of commonly used vegetation indexes according to the unmanned aerial vehicle image band, and determine the vegetation index and corresponding threshold with the highest classification accuracy of the BLB damaged rice;
[0009] S3, superimpose the classification result of the unmanned aerial vehicle multispectral image with the Sentinel-2 pixel space, calculate the area proportion and vegetation index of the disease area in each Sentinel-2 pixel, and construct a BLB damaged rice severity index;
[0010] S4, taking the Sentinel-2 image reflectivity and vegetation index as the independent variable, and taking the BLB damaged rice severity index as the dependent variable, determine the optimal model through multi-model comparison and cross validation;
[0011] S5, using the optimal model and the Sentinel-2 image, draw a spatial distribution map of the severity of the BLB damaged rice in the study area and divide it into grades.
[0012] As preferred, in S3, the calculation formula of the BLB damaged rice severity index is:
[0013]
[0014] Wherein, BLB_DSI is the BLB damaged rice severity index, N BLB represents the number of unmanned aerial vehicle pixels marked as BLB damaged rice in the Sentinel-2 pixel, N total represents the number of all unmanned aerial vehicle pixels in the Sentinel-2 pixel, VI all represents the normalized vegetation index value of all unmanned aerial vehicle pixels in a Sentinel-2 pixel.
[0015] As preferred, S2 includes:
[0016] S201, according to the ground investigation data and visual interpretation result, select the sample area of healthy rice and BLB damaged rice from the unmanned aerial vehicle multispectral image;
[0017] S202, using the spectral index threshold method, combined with a plurality of commonly used vegetation indexes, classify the healthy rice and BLB damaged rice in the unmanned aerial vehicle multispectral image;
[0018] S203, evaluate the classification effect of a plurality of commonly used vegetation indexes, and determine the vegetation index and corresponding threshold with the highest classification accuracy of the BLB damaged rice.
[0019] As preferred, S3 includes:
[0020] S301, spatially superimpose the unmanned aerial vehicle classification result with the pixels of the Sentinel-2 image, to ensure that each Sentinel-2 pixel contains several unmanned aerial vehicle pixels;
[0021] S302, count the number of unmanned aerial vehicle pixels affected by the white leaf blight in each Sentinel-2 pixel, and compare with the number of all rice pixels in the pixel, to calculate the proportion of the white leaf blight damage area; at the same time, calculate the average normalized vegetation index value of all unmanned aerial vehicle pixels in each Sentinel-2 pixel, to construct the white leaf blight damage rice severity index.
[0022] As preferred, S4 comprises:
[0023] S401, extract the band reflectance and vegetation index of the Sentinel-2 pixel as the independent variable, and the white leaf blight damage rice severity index as the dependent variable, to construct a sample set;
[0024] S402, use ten-fold cross-validation and grid search to determine the optimal model in the RF model, SVM model and MLR model.
[0025] In a second aspect, a star-ground multi-source data collaborative rice white leaf blight severity large-area mapping system is provided, for executing the method of any one of the first aspect, comprising:
[0026] An acquisition module is configured to acquire unmanned aerial vehicle multi-spectral images and Sentinel-2 images and perform preprocessing;
[0027] A first determination module is configured to determine healthy and white leaf blight damaged rice sample points in the unmanned aerial vehicle multi-spectral images in combination with ground survey data and visual interpretation; calculate a plurality of commonly used vegetation indexes according to the unmanned aerial vehicle image bands, to determine the vegetation index with the highest classification accuracy of rice white leaf blight and the corresponding threshold value;
[0028] A construction module is configured to spatially superimpose the unmanned aerial vehicle multi-spectral image classification result with the Sentinel-2 pixel, to count the area proportion of the disease area and the vegetation index in each Sentinel-2 pixel, and to construct the white leaf blight damage rice severity index;
[0029] A second determination module is configured to use the Sentinel-2 image reflectance and vegetation index as the independent variable, and the white leaf blight damage rice severity index as the dependent variable, to determine the optimal model through multi-model comparison and cross-validation;
[0030] A drawing module is configured to use the optimal model and the Sentinel-2 image to draw the spatial distribution map of the severity of the rice white leaf blight in the study area and to perform grade division.
[0031] In a third aspect, a computer storage medium is provided, and the computer storage medium stores a computer program; the computer program, when running on a computer, causes the computer to execute the method of any one of the first aspect.
[0032] In a fourth aspect, an electronic device is provided, and the electronic device comprises:
[0033] a memory for storing a computer program;
[0034] a processor for executing the computer program to implement the method of any one of the first aspect.
[0035] The present application has the following beneficial effects:
[0036] 1. The present application fuses high-resolution images of unmanned aerial vehicles and Sentinel-2 satellite images to construct a white leaf blight damage severity index of rice, and combines the comparative optimization of support vector machine (SVM), random forest (RF) and multiple linear regression (MLR) models to realize the accurate inversion and spatial mapping of disease severity. This method innovatively solves key technical problems such as multi-source remote sensing data cross-scale fusion, disease feature quantitative extraction and large-scale inversion model optimization, has the characteristics of high monitoring accuracy and wide spatial coverage, and provides scientific basis and technical support for accurate monitoring and prevention and control decision of rice diseases.
[0037] 2. The present application first realizes large-scale, high-precision quantitative inversion and mapping of rice white leaf blight severity based on multi-scale remote sensing data of star and ground; and comprehensively considers the damage area and damage intensity of BLB within a satellite pixel, and the proposed BLB_DSI index can accurately represent the real damage degree of white leaf blight at the Sentinel-2 pixel scale, and can more effectively reflect the disease situation; in addition, the present application also proposes a research framework for large-scale and accurate inversion and mapping of rice white leaf blight severity based on multi-scale cooperation of ground, unmanned aerial vehicle and satellite, and also has the potential to expand to other crop disease monitoring, providing new technical means and method ideas for the field of agricultural disease and pest monitoring. BRIEF DESCRIPTION OF DRAWINGS
[0038] Figure 1 Flowchart of the method for large-area mapping of rice white leaf blight severity based on multi-source data cooperation of star and ground;
[0039] Figure 2 Unmanned aerial vehicle spectral index box plot of white leaf blight and healthy rice samples calculated based on ground and visual interpretation samples;
[0040] Figure 3 Classification result diagram of white leaf blight and healthy rice based on unmanned aerial vehicle multispectral;
[0041] Figure 4 Scatter density and corresponding accuracy plot between BLB_DSI estimated for RF, SVM and MLR and reference BLB_DSI;
[0042] Figure 5 Results of white leaf blight severity mapping based on Sentinel-2 data. DETAILED DESCRIPTION
[0043] The application will be further described below in connection with the embodiments. The following description of the embodiments is only to help understand the application. It should be noted that for ordinary people in the technical field, without departing from the principles of the application, the application can be modified, and these improvements and modifications also fall within the protection scope of the claims of the application.
[0044] Example 1:
[0045] As an embodiment, the application provides a large-area mapping method for rice white leaf blight severity based on star-sky ground multi-source data cooperation, as shown in Figure 1 The method comprises the following steps:
[0046] S1, acquiring unmanned aerial vehicle multispectral images and Sentinel-2 images and performing pretreatment.
[0047] Specifically, acquiring unmanned aerial vehicle multispectral data comprises: acquiring data by using a DJIP4 Multispectral camera with 5 bands, the spatial resolution of the acquired image is 6-10.6 cm (different flight heights have different spatial resolutions), and the center wavelength of the bands is blue light (450 nm), green light (560 nm), red light (650 nm), red edge (730 nm), and near-infrared light (840 nm). DJI Terra software (version 3.7.6) is used to perform image stitching, radiation correction, and geometric correction on the multispectral image. Then, through visual interpretation, the non-rice area in the unmanned aerial vehicle multispectral image is masked.
[0048] Acquiring Sentinel-2 satellite data comprises: downloading Sentinel-2 image data (orbit number: T51RXV) matching the acquisition time (October 15 and 17, 2023) of the research area on the Google Earth Engine platform. The downloaded Sentinel-2 Level-2A data is pretreated by cloud masking, band synthesis, and tiling.
[0049] S2, combining ground investigation data and visual interpretation to determine healthy and white leaf blight damaged rice sample points in the unmanned aerial vehicle multispectral image; calculating a plurality of commonly used vegetation indices according to the bands of the unmanned aerial vehicle image, and determining the vegetation index with the highest classification accuracy of rice white leaf blight and the corresponding threshold (for example,Figure 2 As shown in FIG. 2B.
[0050] For example, based on the ground survey data and visual interpretation results, the classification accuracy of healthy rice and rice area affected by white leaf blight is the highest when NDVI is equal to 0.5, and the classification accuracy reaches 0.989.
[0051] Specifically, S2 comprises:
[0052] S201, according to the ground survey data and visual interpretation results, sample areas of healthy rice and white leaf blight affected rice are selected from the unmanned aerial vehicle multi-spectral image. For example, 238252 healthy rice samples and 289286 white leaf blight affected rice samples are obtained.
[0053] S202, using a spectral index threshold method, combining a plurality of commonly used vegetation indices, the healthy rice and white leaf blight affected rice in the unmanned aerial vehicle multi-spectral image are classified.
[0054] As shown in Table 1, seven common spectral indices are used in the embodiment of the application.
[0055] Table 1 Spectral index information for extracting white leaf blight affected rice in UAV image
[0056]
[0057] S203, the classification effects of a plurality of commonly used vegetation indices are evaluated, and the vegetation index and the corresponding threshold value with the highest classification accuracy of rice white leaf blight are determined.
[0058] Specifically, when classifying the unmanned aerial vehicle image, 70% of the healthy rice and white leaf blight affected rice samples are selected as the training set, and the remaining 30% are classified as the validation set for classification effect evaluation. The classification results are evaluated by overall accuracy (OA), precision, recall (Recall) and F1 score. The results show that the classification effect using NDVI (threshold value set to 0.5) is the best, the NDVI value of healthy rice is greater than 0.5, the NDVI value of white leaf blight affected rice is less than 0.5, and the healthy rice and the area affected by the disease are effectively distinguished (as shown in FIG. 2B), and the classification accuracy OA reaches 0.989. Figure 3
[0059] S3, the classification result of the unmanned aerial vehicle multi-spectral image is superimposed with the Sentinel-2 pixel space, the area proportion and vegetation index of the disease area in each Sentinel-2 pixel are counted, and the severity index of white leaf blight affected rice is constructed.
[0060] S4, taking the Sentinel-2 image reflectivity and the vegetation index as the independent variables, and taking the rice white leaf blight damage severity index as the dependent variable, determining the optimal model through multi-model comparison and cross-validation.
[0061] S5, using the optimal model and the Sentinel-2 image, drawing a spatial distribution map of the rice white leaf blight severity in the study area and performing grade division.
[0062] Embodiment 2:
[0063] Based on embodiment 1, the present application embodiment 2 provides a more specific large-area mapping method for rice white leaf blight severity based on star-ground multi-source data collaboration, including:
[0064] S1, obtaining the unmanned aerial vehicle multi-spectral image and the Sentinel-2 image and performing preprocessing.
[0065] S2, combining the ground survey data and visual interpretation to determine the healthy and white leaf blight damaged rice sample points in the unmanned aerial vehicle multi-spectral image; calculating a plurality of commonly used vegetation indexes according to the unmanned aerial vehicle image band, and determining the vegetation index with the highest classification accuracy of rice white leaf blight and the corresponding threshold.
[0066] S3, spatially superimposing the unmanned aerial vehicle multi-spectral image classification result and the Sentinel-2 pixel to calculate the area proportion and the vegetation index of the disease area in each Sentinel-2 pixel, and constructing the white leaf blight damaged rice severity index.
[0067] Specifically, in each Sentinel-2 pixel, the number of unmanned aerial vehicle pixels of healthy rice and white leaf blight damaged rice is counted, and the corresponding average vegetation index value (NDVI in the present application embodiment) is calculated. Through the white leaf blight damage area proportion and the damage intensity in each satellite pixel, the white leaf blight severity index (BLB_DSI) of each Sentinel-2 pixel is constructed. The specific process is as follows:
[0068] S301, spatially superimposing the unmanned aerial vehicle classification result and the Sentinel-2 image pixel to ensure that each Sentinel-2 pixel contains a plurality of unmanned aerial vehicle pixels.
[0069] This step is realized by the fishing net tool in R language, which ensures that each Sentinel-2 pixel contains a plurality of unmanned aerial vehicle pixels, thereby facilitating the subsequent accurate extraction of the rice disease information in each pixel.
[0070] S302, the number of unmanned aerial vehicle pixels affected by white leaf blight in each Sentinel-2 pixel (i.e. the infection area) is counted and compared with the number of all rice pixels in the pixel, so as to calculate the proportion of white leaf blight damage area; at the same time, the average normalized vegetation index value of all unmanned aerial vehicle pixels in each Sentinel-2 pixel is calculated, and a white leaf blight damage rice severity index is constructed.
[0071] In S302, in order to quantify the infection intensity, the NDVI value of healthy rice is defined as the theoretical maximum value (i.e. 1), and the NDVI value of white leaf blight damaged rice is compared with the NDVI value of healthy rice to calculate the deviation as an index of infection intensity. Finally, by combining the damage area and damage intensity of white leaf blight damaged rice, a white leaf blight damage rice severity index (BLB_DSI) is constructed, and the calculation formula is:
[0072]
[0073] BLB_DSI is the severity of white leaf blight damage rice in a Sentinel-2 pixel, N BIB represents the number of unmanned aerial vehicle pixels marked as white leaf blight damage rice in the pixel, N total represents the number of all unmanned aerial vehicle pixels in the Sentinel-2 pixel. VI all represents the NDVI value of all unmanned aerial vehicle pixels in a Sentinel-2 pixel.
[0074] S4, taking the reflectivity and vegetation index of Sentinel-2 image as independent variables and the white leaf blight damage rice severity index as dependent variable, comparing the accuracy of support vector machine (SVM), random forest (RF) and multiple linear regression (MLR) models, determining the optimal model by cross-validation and using it for large-scale white leaf blight damage severity mapping.
[0075] S4 includes:
[0076] S401, extracting the band reflectivity and vegetation index (as shown in Table 1) of Sentinel-2 pixel as independent variables, and constructing a sample set with the white leaf blight damage rice severity index as dependent variable.
[0077] In addition, all variables are Z-score standardized, and divided into training set (13,764) and validation set (5,899) according to the ratio of 7:3.
[0078] S402, using ten-fold cross-validation and grid search to determine the optimal model among RF model, SVM model and MLR model.
[0079] Specifically, the R language caret package is used for model training and hyperparameter optimization. Ten-fold cross-validation and grid search are used to determine the optimal parameters: RF model, SVM model, and MLR model. The optimal inversion model of the severity of BLB of rice is determined by the validation coefficient of determination, mean absolute error, and root mean square error. Among them, RF has the best validation accuracy, R 2 is 0.91, RMSE is 8.58%, and MAE is 5.48%. The second is SVM, and the validation accuracy of MLR is the lowest, R 2 is 0.816.
[0080] S5, using the optimal model and Sentinel-2 image, a spatial distribution map of the severity of BLB of rice in the study area is drawn and classified.
[0081] Specifically, S5 includes:
[0082] S501, applying the optimal model to the Sentinel-2 image of Ningbo City, inputting the band reflectance and vegetation index, and outputting BLB_DSI, the spatial distribution of the severity of BLB of rice in Ningbo City.
[0083] S502, in order to more clearly show the severity of BLB, the disease severity is classified (healthy, mild, moderate, and severe), and a spatial distribution map (as shown in Figure 5 ) is generated. Among them, according to the distribution of BLB_DSI prediction value, interval method is used to divide BLB_DSI value into five levels (as shown in Table 2), and the threshold is:
[0084] T a ≤ BLB_DSI < T b
[0085] Table 2
[0086]
[0087] Among them, BLB_DSI is the severity index of BLB of rice, T a and T b are the maximum and minimum values of BLB_DSI, respectively. In this embodiment, T1, T2, T3, T4, T5 and T6 are 0, 0.2, 0.4, 0.6, 0.8 and 1, respectively.
[0088] It should be noted that the parts in this embodiment that are the same as or similar to those in Embodiment 1 can be mutually referenced, and will not be repeated in this application.
[0089] Embodiment 3:
[0090] Based on embodiment 2, the application embodiment 3 provides a star-ground multi-source data collaborative system for large-area mapping of rice bacterial leaf blight severity, including:
[0091] An acquisition module is configured to acquire and pre-process the unmanned aerial vehicle multi-spectral image and the Sentinel-2 image.
[0092] A first determination module is configured to determine healthy and bacterial leaf blight damaged rice sample points in the unmanned aerial vehicle multi-spectral image in combination with ground investigation data and visual interpretation, and to determine a vegetation index with the highest classification accuracy of rice bacterial leaf blight and a corresponding threshold value according to the unmanned aerial vehicle image band.
[0093] A construction module is configured to superimpose the unmanned aerial vehicle multi-spectral image classification result and the Sentinel-2 pixel space, to calculate the area proportion and vegetation index of the disease area in each Sentinel-2 pixel, and to construct a rice bacterial leaf blight severity index.
[0094] A second determination module is configured to take the Sentinel-2 image reflectivity and the vegetation index as independent variables, to take the rice bacterial leaf blight severity index as a dependent variable, and to determine an optimal model through multi-model comparison and cross-validation.
[0095] A drawing module is configured to draw a spatial distribution map of the rice bacterial leaf blight severity in the research area and to perform grade division by using the optimal model and the Sentinel-2 image.
[0096] It should be noted that the system provided in this embodiment is a system corresponding to the method provided in embodiment 2, and therefore, the same or similar parts in this embodiment and embodiment 2 can be mutually referenced, and will not be described herein again.
[0097] In summary, the present application proposes a method for mapping the severity of rice bacterial leaf blight (BLB) based on the synergy of multi-source data from space and ground. First, unmanned aerial vehicle (UAV) multispectral data and Sentinel-2 satellite imagery are acquired and preprocessed. The UAV data is processed through image stitching, radiometric correction, and geometric correction, while the Sentinel-2 imagery is processed through cloud masking and band synthesis to ensure data quality. Then, various vegetation indices (such as NDVI and EVI) are calculated, and the classification threshold for healthy rice and BLB-affected rice is determined based on ground survey data to distinguish between healthy and affected areas. Next, the UAV classification results are spatially overlaid with Sentinel-2 pixels, and the proportion of disease area and the severity of damage (represented by the size of the vegetation index value) within each satellite pixel are calculated to construct a BLB disease severity index (BLB_DSI). Finally, random forest (RF), support vector machine (SVM), and multiple linear regression (MLR) models are used to model the reflectance and vegetation index of Sentinel-2 as independent variables and the BLB_DSI as the dependent variable. Cross-validation is used to optimize the optimal model, and a large-scale spatial distribution map of rice BLB severity is generated. This method can effectively identify the spatial distribution and severity of rice BLB, providing a scientific basis for agricultural disease monitoring and prevention decision-making, and has important practical application value.
Claims
1. A method for mapping the severity of rice bacterial leaf blight on a large scale by coordinating multi-source data of the sky and land, characterized in that, The method comprises the following steps: S1, acquiring and preprocessing unmanned aerial vehicle multi-spectral images and Sentinel-2 images; S2, determining healthy and white leaf blight damaged rice sample points in the unmanned aerial vehicle multi-spectral images in combination with ground investigation data and visual interpretation; calculating a plurality of commonly used vegetation indexes according to unmanned aerial vehicle image bands to determine a vegetation index with the highest classification accuracy of rice white leaf blight and a corresponding threshold value; S3, superimposing the unmanned aerial vehicle multi-spectral image classification result and the Sentinel-2 pixel space to calculate the area proportion of the disease area in each Sentinel-2 pixel and the vegetation index, and constructing a white leaf blight damaged rice severity index; S3 comprises the following steps: S301, superimposing the unmanned aerial vehicle classification result and the Sentinel-2 image pixel space to ensure that each Sentinel-2 pixel contains a plurality of unmanned aerial vehicle pixels; S302, counting the number of unmanned aerial vehicle pixels affected by the white leaf blight in each Sentinel-2 pixel, and comparing it with the number of all rice pixels in the pixel to calculate the white leaf blight damage area proportion; at the same time, calculating the average normalized vegetation index value of all unmanned aerial vehicle pixels in each Sentinel-2 pixel to construct a white leaf blight damaged rice severity index; S4, taking the Sentinel-2 image reflectivity and the vegetation index as the independent variable, and taking the white leaf blight damaged rice severity index as the dependent variable, determining the optimal model through multi-model comparison and cross-validation; S5, using the optimal model and the Sentinel-2 image to draw a spatial distribution map of the severity of the white leaf blight of rice in the study area and to divide the grades.
2. The method for large-area mapping of rice bacterial leaf blight severity using multi-source data from the starry sky and ground as described in claim 1, is characterized in that... In S3, the calculation formula of the white leaf blight damaged rice severity index is: where BLB DSI is the severity index of BLB damage to rice, N BLB represents the number of drone pixels within the Sentinel-2 pixel that are marked as having BLB damage to rice, N total represents the total number of drone pixels within the Sentinel-2 pixel, represents the normalized difference vegetation index value for all drone pixels within a Sentinel-2 pixel.
3. The method according to claim 2, wherein S2 The method comprises the following steps: S201, selecting healthy rice and white leaf blight damaged rice sample areas from the unmanned aerial vehicle multi-spectral images according to the ground investigation data and the visual interpretation result; S202, classifying healthy rice and white leaf blight damaged rice in the unmanned aerial vehicle multi-spectral images by using the spectral index threshold method in combination with a plurality of commonly used vegetation indexes; S203, evaluating the classification effect of a plurality of commonly used vegetation indexes to determine the vegetation index with the highest classification accuracy of rice white leaf blight and the corresponding threshold value.
4. The method according to claim 3, wherein S4 The method comprises the following steps: S401, extracting the band reflectivity and the vegetation index of the Sentinel-2 pixel as the independent variable, and taking the white leaf blight damaged rice severity index as the dependent variable to construct a sample set; S402, determining the optimal model in the RF model, the SVM model and the MLR model by using ten-fold cross-validation and grid search.
5. A system for mapping the severity of rice bacterial leaf blight on a large scale by using multi-source data in synergy with the starry sky, characterized by, The device for executing the method of any one of claims 1 to 4 comprises: an acquisition module configured to acquire and preprocess unmanned aerial vehicle multi-spectral images and Sentinel-2 images; a first determination module configured to determine healthy and white leaf blight damaged rice sample points in the unmanned aerial vehicle multi-spectral images in combination with ground investigation data and visual interpretation; and calculate a plurality of commonly used vegetation indexes according to unmanned aerial vehicle image bands to determine a vegetation index with the highest classification accuracy of rice white leaf blight and a corresponding threshold value; The constructing module is configured to superimpose the unmanned aerial vehicle multi-spectral image classification result and a Sentinel-2 pixel space, count an area proportion of a disease area and a vegetation index in each Sentinel-2 pixel, and construct a rice white leaf blight damage severity index; The second determining module is configured to take the Sentinel-2 image reflectivity and the vegetation index as independent variables, take the rice white leaf blight damage severity index as a dependent variable, and determine an optimal model through multi-model comparison and cross-validation; The drawing module is configured to draw a spatial distribution map of the rice white leaf blight severity in the research area and perform grade division by using the optimal model and the Sentinel-2 image.
6. A computer storage medium, characterized in that The computer storage medium stores a computer program; and the computer program, when running on a computer, causes the computer to execute the method in any one of claims 1 to 4.
7. An electronic device, comprising: The computer program product comprises: a memory configured to save the computer program; and a processor configured to execute the computer program to implement the method in any one of claims 1 to 4.