Multi-data fusion large-area rice disease and pest remote sensing monitoring method and system

Abstract

The invention belongs to the technical field of rice disease and insect pest monitoring applied to remote sensing, and provides a multi-data fused large-area rice disease and insect pest remote sensing monitoring method and system for solving the problems that in an existing monitoring method, multi-source data fusion application is insufficient, scales are different, and large-area disease and insect pest monitoring is difficult to achieve. S2, carrying out sample category balance processing on the ground-unmanned aerial vehicle sample set, and extracting a rice pest and disease damage area on the unmanned aerial vehicle image; s3, performing multi-scale data fusion on the initial multiple data; s4, performing integrated modeling training of the large-area rice disease and insect pest recognition model and obtaining a final monitoring result; the system comprises a ground sampling module, an unmanned aerial vehicle monitoring module, a data fusion module and a satellite monitoring module, precise monitoring of large-area rice diseases and insect pests under different farmland habitat conditions is achieved, and the problems that an existing method is small in coverage range, low in monitoring efficiency and insufficient in data fusion application are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of rice pest and disease monitoring in remote sensing applications. Background Art

[0002] Rice pests and diseases are important factors affecting rice yield and quality, posing a serious threat to agricultural production, economic benefits, food security, and ecological security. There are more than 90 common rice pests and diseases in China. Among them, pests and diseases such as rice planthoppers, rice leaf folders, striped stem borers, rice sheath blight, and rice blast are in a severe outbreak situation, causing a yield loss of millions of tons every year.

[0003] Remote sensing technology analyzes the properties, characteristics, and states of ground objects by receiving their radiation information. It has advantages such as a wide coverage area, large amount of information, strong timeliness, and low cost. It can provide quantitative information on the spatio-temporal changes of farmland crops and is suitable for large-scale rice pest and disease monitoring. With the continuous development and popularization of sensor technology and unmanned aerial vehicle platforms, remote sensing monitoring has become an important part of the rice pest and disease monitoring system.

[0004] The invention patent with the publication number CN112507770B discloses "a method and system for identifying rice pests and diseases". After collecting the images of rice pests and diseases to be identified in the paddy field, a trained deep learning identification model is called to identify the images of rice pests and diseases to be identified, and the identification results are output. According to the identification results, it is judged whether the rice to be identified has pests and diseases. If so, an alarm prompt is issued, and the names of the pests and diseases and the corresponding control measures are output.

[0005] The invention patent with the publication number CN111537668B discloses "a method and device for remote sensing monitoring of crop pests and diseases based on meteorological satellite data". The method includes obtaining meteorological data and processing it to obtain key meteorological elements; calculating the daily data of key meteorological elements according to Beijing time; establishing a corresponding meteorological evaluation model for the incidence index of pests and diseases using mathematical statistics methods; calculating the incidence index of pests and diseases with a ten-day period; and performing masking using the distribution data of the target crop to generate a thematic map.

[0006] The invention patent with the publication number CN118624540B discloses "a method for pest and disease monitoring based on the collaboration of remote sensing and meteorological data". Remote sensing images of multiple sub-regions in the target area are obtained and preprocessed; the sub-regions with pests and diseases are identified using a pixel unit matching algorithm; the historical meteorological data and the spectral characteristic wavelengths of the corresponding remote sensing images of the pest and disease sub-regions are obtained; a pest and disease outbreak prediction model is established, and the meteorological data and the spectral characteristic wavelengths of the remote sensing images are input into the pest and disease outbreak prediction model for pest and disease monitoring.

[0007] However, the above remote sensing monitoring methods for pests and diseases still have the following deficiencies: (1) Multi-source data is not fully integrated and applied. Although meteorological data can reflect the environmental characteristics of pest development or pathogen transmission, the integrated application of ground precise sampling values, high-resolution images of unmanned aerial vehicles (UAVs), and large-scale images of satellites is also important in pest and disease monitoring, which can provide more detailed information on the spatial distribution of pests and diseases; (2) It is difficult to achieve large-area pest and disease monitoring. The existing remote sensing monitoring methods mostly obtain the pest and disease identification results of images at individual positions in the field, and the samples between different scales are not effectively migrated and used, and the coverage of pest and disease monitoring has not been extended to large areas; (3) The single modeling method has limitations, resulting in low accuracy of pest and disease monitoring. Summary of the Invention

[0008] In view of the insufficient integration and application of multi-source data, different scales, and the difficulty in achieving large-area pest and disease monitoring in the existing monitoring methods, and the inability to fully utilize the advantages of the all-round three-dimensional monitoring of remote sensing technology, the present invention proposes a "remote sensing monitoring method and system for rice pests and diseases in large areas with multi-data fusion".

[0009] The remote sensing monitoring method for rice pests and diseases in large areas with multi-data fusion, as Figure 1 shown, includes the following steps:

[0010] S1. Initial multi-data collection, including ground data sampling: A mobile verification APP and an Internet of Things monitoring device are respectively used to collect rice field disease information and rice field pest information to form ground data; UAV image data sampling: Set the flight route and altitude of the UAV within a small area, determine the image shooting method, obtain the multi-spectral UAV images of the sample area, mark the ground data on the UAV images and collect the values of each band to form a ground-UAV sample set; and satellite image data sampling;

[0011] S2. Perform sample category balance processing on the ground-UAV sample set, and extract the rice pest and disease areas on the UAV images;

[0012] S3. The ground data in the initial multi-data is point data, the UAV image data is high-resolution surface data, and the satellite image data is medium-low resolution surface data. Perform data fusion between different scales, specifically:

[0013] S31. Unify the data coordinate system;

[0014] S32. Register the rice pest and disease areas on the UAV images extracted in S2 with the satellite image data;

[0015] S33. Calculate the disease incidence index DI of rice pests and diseases for different satellite pixels,

[0016]

[0017] Among them, S i is the area of each UAV pixel with rice pests and diseases within the satellite pixel range, and S all is the total area of this satellite pixel.

[0018] S34. Each satellite pixel within the rice planting range has a pest and disease occurrence degree index DI and multiple characteristic bands. The DI value contains information at the ground sampling scale and the UAV scale, and the characteristic bands contain information at the satellite scale, thereby jointly constructing a multi-scale data fusion sample set and verifying the effectiveness of the multi-scale data fusion sample set;

[0019] S4. Use the multi-scale data fusion sample set constructed in S3 to perform integrated modeling training for the large-area rice pest and disease identification model and obtain the final monitoring result:

[0020] S41. Divide the multi-scale data fusion sample set into multiple subsets. For each subset, use different base models for iterative training and prediction. Each base model will train the subset and generate a prediction result for the data;

[0021] S42. Use the prediction results of each base model as new features to form a new training data set to train the meta-model; use the trained meta-model to predict the test data. The meta-model combines and weights according to the prediction results of the base models to generate a large-area rice pest and disease identification model. After preprocessing the satellite image data of the large area, input it into the large-area rice pest and disease identification model to start prediction, and obtain the large-area rice pest and disease remote sensing monitoring result according to the pest and disease occurrence degree index of different pixels in the prediction result.

[0022] The large-area rice pest and disease remote sensing monitoring system with multi-data fusion, as Figure 2 shown, includes a ground sampling module, which carries a mobile verification APP and Internet of Things monitoring devices. The mobile verification APP is used for sampling and storing ground data, and the Internet of Things monitoring devices are used for continuously monitoring the distribution quantity and population change dynamics of field pests within the surrounding range;

[0023] A UAV monitoring module, which is used to achieve sample category balance processing and extract the rice pest and disease areas on the UAV images and transmit them to the data fusion module;

[0024] A data fusion module, which is used to perform data fusion between multi-scales of ground data, UAV image data, and satellite image data to form a multi-scale data fusion sample set and verify the effectiveness of the multi-scale data fusion sample set;

[0025] The satellite monitoring module is used to perform integrated modeling training of the large - area rice pest and disease identification model based on the constructed multi - scale data fusion sample set, and call the trained large - area rice pest and disease identification model to predict the satellite image data of the large area. According to the rice pest and disease occurrence degree index of different pixels in the prediction results, the remote sensing monitoring results of large - area rice pests and diseases are obtained.

[0026] Technical effects:

[0027] The present invention proposes a multi - data - fusion remote sensing monitoring method for large - area rice pests and diseases, which realizes the accurate monitoring of large - area rice pests and diseases under different farmland habitat conditions, and solves the problems of small coverage, low monitoring efficiency, and insufficient data fusion application in the existing rice pest and disease monitoring. Aiming at the single - modeling problem of the existing monitoring methods, modeling methods such as random forest, support vector machine, and KNN each have their own advantages and limitations. The present invention comprehensively utilizes the advantages of different models through machine learning integrated modeling to improve the accuracy of pest and disease monitoring. And comprehensively utilizes multi - modal and all - round remote sensing monitoring technical means to construct a multi - data - fusion multi - source remote sensing monitoring system for rice pests and diseases. By comprehensively applying the above - mentioned multi - source remote sensing monitoring framework, the accurate monitoring of large - area rice pests and diseases is finally realized.

[0028] Through the present invention, pest and disease monitoring is carried out on a rice planting area in a certain place in Guangxi. The root - mean - square error RMSE and R 2 are used to evaluate the performance of the large - area rice pest and disease identification model.

[0029]

[0030] where y i is the actual value, is the predicted value, is the average of the actual values, n is the number of samples, and appropriate parameter tuning and verification are carried out to avoid overfitting and improve the stability of the model.

[0031] The following table shows the accuracy indicators of this pest and disease remote sensing monitoring method on the validation set. Figure 3 It is a visible view of the spatial distribution result of the monitored rice pests and diseases, and can accurately determine the location and corresponding occurrence degree of the rice pests and diseases according to this result.

[0032] RMSE <![CDATA[R 2 > 0.0928 0.7419

[0033] The RMSE value is 0.0928, indicating that the average error between the predicted value and the actual value is small. The R 2 value is 0.7419, indicating that the fitting effect of the large - area rice pest and disease identification model is good, and can truly quantify the incidence degree of rice pests and diseases. Description of the Drawings

[0034] Figure 1 This is the overall flowchart of the remote sensing monitoring method for rice pests and diseases in large areas with multi-data fusion of the present invention.

[0035] Figure 2 This is the schematic block diagram of the remote sensing monitoring system for rice pests and diseases in large areas with multi-data fusion of the present invention.

[0036] Figure 3 This is the viewable result of the spatial distribution of rice pests and diseases monitored by the embodiment of the present invention. Detailed implementation manners

[0037] In order to better understand the technical solution of the present invention, the embodiments provided by the present invention will be described in detail below with reference to the accompanying drawings, but the implementation manners of the present invention are not limited thereto.

[0038] As Figure 1 shown, the remote sensing monitoring method for rice pests and diseases in large areas with multi-data fusion includes the following steps:

[0039] S1. Initial multi-data collection, including ground data sampling: The mobile verification APP and the Internet of Things monitoring devices are respectively used to collect rice field disease information and rice field pest information to form ground data. Since the occurrence areas and positions of diseases are uncertain, field workers need to independently determine the sampling positions according to the on-site occurrence degree to improve the flexibility of sampling; UAV image data sampling: Set the flight routes and altitudes of the UAVs within a small area, determine the image shooting method, obtain the multi-spectral UAV images of the sample area, mark the ground data on the UAV images, and collect the values of each band of the UAV images within a certain buffer area where the sample points are located to form a ground-UAV sample set; and satellite image data sampling.

[0040] Further, the ground data format in S1 is a list, and the attributes include the longitude and latitude of the point, the name of the pest and disease, and the occurrence degree.

[0041] S2. Perform sample category balance processing on the ground-UAV sample set, and extract the rice pest and disease areas on the UAV images;

[0042] Further, the sample category balance processing is specifically as follows:

[0043] S211. Count the number of samples under different pest and disease occurrence degrees. When there are M sample data in the N-classified sample set, define that the number of a certain type of sample is less than 0.6×M / N as the minority class sample, and the number of a certain type of sample is higher than 1.4×M / N as the majority class sample;

[0044] S212. Use the SMOTE method to generate synthetic samples for the minority class samples. For each minority class sample, the SMOTE algorithm identifies its K nearest neighbor samples in the feature space, selects a random neighbor, and calculates the difference between the feature vector of the minority sample and this neighbor. Then, multiply the difference by a random number between 0 and 1 and add it to the feature vector of the minority sample to create a new synthetic sample. Repeat this process until the number of the minority class samples reaches the range of [0.9×M / N, 1.1×M / N].

[0045] S213. Use the ClusterCentroids method to reduce the number of samples for the majority class samples. For each type of majority class samples, use the K-Means algorithm for clustering to obtain K cluster centers, and use the feature space coordinates of these cluster centers as new samples to replace the original majority class samples, so that the number of the majority class samples reaches the range of [0.9×M / N, 1.1×M / N].

[0046] Further, the specific method for extracting the rice pest and disease area on the UAV image is as follows:

[0047] S221. Calculate the band feature means and vegetation index feature means of the minority class and majority class samples.

[0048] S222. Compare the differences in feature means between different class samples, and select the feature dimension with the largest relative difference in feature means as the threshold segmentation dimension.

[0049] S223. For the single-band feature image of the threshold segmentation dimension, overlay the vector of the rice planting area to determine the distribution range of the rice, and then use the adaptive threshold segmentation method to segment the rice area on the UAV image into the area without pests and diseases and the area with pests and diseases, so as to extract the rice pest and disease area.

[0050] S3. The ground data in the initial multi-data is point data, the UAV image data is high-resolution surface data, and the satellite image data is medium-low resolution surface data. Perform data fusion between multiple scales, specifically as follows:

[0051] S31. Unify the data coordinate system. Preferably, the unified data coordinate system is the WGS84 geographic coordinate system, and transform the data under other coordinate systems to this coordinate system.

[0052] S32. Register the rice pest and disease area on the UAV image extracted in S2 with the satellite image data to ensure accurate spatial position correspondence.

[0053] S33. Calculate the rice pest and disease occurrence degree index DI of different satellite pixels.

[0054]

[0055] Among them, S i is the area of each UAV pixel with rice pests and diseases within the satellite pixel range, and S all is the total area of the satellite pixel. The value range of DI is [0, 1], and the larger its value indicates the more serious the occurrence degree of pests and diseases.

[0056] S34. Each satellite pixel within the rice planting range has a pest and disease occurrence degree index DI and multiple characteristic bands. The DI value contains information at the ground sampling scale and the UAV scale, and the characteristic bands contain information at the satellite scale, thus jointly constructing a multi-scale data fusion sample set and testing the effectiveness of the multi-scale data fusion sample set;

[0057] Furthermore, testing the effectiveness of the multi-scale data fusion sample set specifically includes:

[0058] S341. Calculate the statistical characteristics of the multi-scale data fusion sample set, including the mean, standard deviation, and confidence interval, verify the outliers, and eliminate the sample values affected by cloud occlusion or ground mixed land cover interference;

[0059] S342. Calculate the correlation between the pest and disease occurrence degree index of the satellite pixel and the occurrence degree at the ground sampling points within the corresponding pixel. If the Pearson correlation coefficient exceeds 0.5, it indicates that the calculation method of the pest and disease occurrence degree in the multi-scale data fusion sample set is reasonable;

[0060] S343. Statistically analyze the spatial positions of the centers of all sampled satellite pixels and calculate their nearest neighbor distances to other centers, and then use the K-S test to determine the difference between the distance empirical distribution function and the theoretical distribution function. If there is a significant difference, it indicates that the sampling positions of the satellite pixels follow the random distribution hypothesis.

[0061] S4. Use the multi-scale data fusion sample set constructed by S3 to perform integrated modeling training for the large-area rice pest and disease identification model and obtain the final monitoring results:

[0062] S41. Use the five-fold cross-validation method to divide the multi-scale data fusion sample set into multiple subsets. For each subset, use different base models for iterative training and prediction. Each base model will train the subset and generate prediction results for the data; these base models are mainly machine learning algorithms, including logistic regression, decision tree, gradient boosting tree, support vector machine, KNN, naive Bayes, etc.

[0063] S42. Use the prediction results of each base model as new features to combine into a new training dataset for training the meta-model; these prediction results serve as supplements to the original features, providing more information for training the meta-model; use the trained meta-model to predict the test data. The meta-model is a random forest model that can learn how to combine the prediction results of the base models to maximize the accuracy of the overall model; the meta-model combines and weights according to the prediction results of the base models to generate a large-area rice pest and disease identification model. After preprocessing the satellite image data of the large area, input it into the large-area rice pest and disease identification model to start the prediction, and obtain the large-area rice pest and disease remote sensing monitoring results based on the rice pest and disease occurrence degree index of different pixels in the prediction results.

[0064] Further, the preprocessing of the satellite image data of the large area includes downloading the satellite image of the large area, and after preprocessing steps such as radiometric calibration, atmospheric correction, geometric correction, cloud and shadow removal, overlay the rice area vector, and crop to obtain the rice satellite image data.

[0065] The multi-data fusion large-area rice pest and disease remote sensing monitoring system, as Figure 2 shown, includes a ground sampling module, which carries a mobile verification APP and Internet of Things monitoring devices. The mobile verification APP is used for sampling and storing ground data, and the Internet of Things monitoring devices are used for continuously monitoring the distribution quantity and population change dynamics of field pests within the surrounding range;

[0066] The UAV monitoring module is used to achieve sample category balance processing and extract the rice pest and disease areas on the UAV images and transmit them to the data fusion module;

[0067] The data fusion module is used to perform multi-scale data fusion between ground data, UAV image data, and satellite image data to form a multi-scale data fusion sample set, and to test the effectiveness of the multi-scale data fusion sample set;

[0068] The satellite monitoring module is used to perform integrated modeling training of the large-area rice pest and disease identification model according to the constructed multi-scale data fusion sample set, and call the trained large-area rice pest and disease identification model to predict the satellite image data of the large area, and obtain the large-area rice pest and disease remote sensing monitoring results based on the rice pest and disease occurrence degree index of different pixels in the prediction results.

[0069] The mobile verification APP is mainly used to collect rice field disease information and consists of a mobile device terminal, a data transmission layer, a background management system, a user service layer, and a GIS service system:

[0070] The mobile device terminal includes smartphones and tablets, which are used as carriers to install the mobile verification APP for on-site ground data collection;

[0071] Data transmission layer: Output the on-site ground data acquired by sensors such as mobile phone or tablet cameras, as well as the longitude and latitude data acquired by the positioning and navigation system, to the background management system through the mobile network;

[0072] Background management system: Responsible for receiving, processing, storing, and analyzing the data transmitted by the data transmission layer;

[0073] User service layer: Set different permissions for different levels of users and open different data;

[0074] GIS service system: Responsible for the management and loading of different layers, as well as the display and publishing of sampling points.

[0075] The sampling method of the Internet of Things monitoring device is mainly used to collect information on rice field pests. It captures pests through a baiting device and can continuously monitor the distribution quantity and population change dynamics of field pests within a certain range around in real time. The Internet of Things monitoring device consists of an insect attracting device, an insect collector, a high-definition camera, a power supply component, a data transmission device, and an AI intelligent recognition system.

[0076] Insect attracting device: Complete the functions of attracting, killing, and dispersing insect bodies of rice field pests by using light, electricity, and chemical insect attractants;

[0077] Insect collector: Responsible for collecting and spreading the killed pest bodies on the display whiteboard for subsequent intelligent recognition and counting, and regularly cleaning and removing the pest bodies;

[0078] High-definition camera: Regularly take pictures of the pest distribution on the whiteboard of the insect collector and transmit the high-definition pictures to the AI intelligent recognition system;

[0079] Power supply component: Obtain energy input through devices such as solar panels to reduce the equipment maintenance cost;

[0080] Data transmission device: Transmit the data collected by the Internet of Things device to the data center, and the communication methods include wireless communication and wired communication;

[0081] AI intelligent recognition system: Automatically recognize and count the taken high-definition pictures based on the AI recognition algorithm for plant diseases and insect pests, and count the types and quantity changes of pests.

[0082] The content not described in detail in this specification belongs to the prior art well-known to those skilled in the art. At the same time, for those of ordinary skill in the art, there will be changes in the specific implementation manners and application scopes according to the idea of the present invention. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A remote sensing monitoring method for rice pests and diseases in a large area based on multi-data fusion, characterized in that: The steps include: S1. Initial multi-data collection, including ground data sampling: mobile verification APP and IoT monitoring equipment are used to collect rice field disease information and rice field pest information respectively to form ground data; drone image data sampling: setting the flight route and altitude of the drone in a small area, determining the image shooting method, obtaining multispectral drone images of the sample area, marking the ground data on the drone images and collecting the values ​​of each band to form a ground-drone sample set; and satellite image data sampling; S2: Perform sample category balancing on the ground-UAV sample set and extract rice pest and disease areas on the UAV image; S3, the ground data in the initial multi-data is point data, the drone image data is high-resolution surface data, and the satellite image data is medium- and low-resolution surface data. Multi-scale data fusion is performed, specifically: S31, unified data coordinate system; S32, registering the rice pest and disease area on the drone image extracted in S2 with the satellite image data; S33, calculate the rice pest and disease occurrence index DI of different satellite pixels, Among them, S i S is the area of ​​each UAV pixel within the range of satellite pixels where rice pests and diseases occur. all is the total area of ​​the satellite pixel, S34. Each satellite pixel in the rice planting area has a pest and disease occurrence index DI and multiple characteristic bands. The DI value contains information on the ground sampling scale and the drone scale, and the characteristic band contains information on the satellite scale, thereby jointly constructing a multi-scale data fusion sample set and testing the effectiveness of the multi-scale data fusion sample set; S4, using the multi-scale data fusion sample set constructed in S3 to conduct integrated modeling training of large-area rice pest and disease identification model and obtain the final monitoring results: S41, dividing the multi-scale data fusion sample set into multiple subsets, for each subset, using a different base model for iterative training and prediction, each base model will train the subset and generate a prediction result for the data; S42. The prediction results of each base model are used as new features and combined into a new training data set to train the meta-model; the trained meta-model is used to predict the test data, and the meta-model is combined and weighted according to the prediction results of the base models to generate a large-area rice pest and disease recognition model, and the satellite image data of the large area is pre-processed and input into the large-area rice pest and disease recognition model to start prediction, and the large-area rice pest and disease remote sensing monitoring results are obtained according to the rice pest and disease occurrence degree index of different pixels in the prediction results.

2. The method for remote sensing monitoring of rice pests and diseases in a large area by multi-data fusion according to claim 1 is characterized in that: The ground data format described in S1 is a list, and the attributes include the latitude and longitude of the point, the name of the pest and disease, and the degree of occurrence.

3. The method for remote sensing monitoring of rice pests and diseases in a large area by multi-data fusion according to claim 1 is characterized in that: The sample category balancing process described in S2 is specifically as follows: S211. Count the number of samples under different degrees of pests and diseases. When there are M sample data in the N-classified sample set, define a class of samples with a number less than 0.6×M / N as minority class samples, and a class of samples with a number greater than 1.4×M / N as majority class samples. S212. Generate synthetic samples using the SMOTE method for minority class samples. For each minority class sample, the SMOTE algorithm identifies its K nearest neighbor samples in the feature space, selects a random neighbor, and calculates the difference between the feature vector of the minority sample and the neighbor; then, multiplies the difference by a random number between 0 and 1 and adds it to the feature vector of the minority sample to create a new synthetic sample. Repeat this process until the number of minority class samples reaches the range of [0.9×M / N,1.1×M / N]. S213. Use the ClusterCentroids method to reduce the number of samples for the majority class samples. For each type of majority class samples, use the K-Means algorithm to cluster them and obtain K cluster centers. Use the feature space coordinates of these cluster centers as new samples to replace the original majority class samples, so that the number of majority class samples reaches the range of [0.9×M / N,1.1×M / N].

4. The method for remote sensing monitoring of rice pests and diseases in a large area by multi-data fusion according to claim 3 is characterized in that: The rice pest and disease areas extracted from the drone image described in S2 are specifically: S221, calculating the band feature mean and vegetation index feature mean of the minority class and majority class samples; S222, comparing the differences in feature means between samples of different categories, and selecting the feature dimension with the largest relative difference in feature means as the threshold segmentation dimension; S223. For the single-band feature image of the threshold segmentation dimension, the rice planting area vector is superimposed to determine the distribution range of rice, and then the rice area on the drone image is segmented into areas without pests and diseases and areas with pests and diseases using the adaptive threshold segmentation method, so as to extract the rice pest and disease areas.

5. The method for remote sensing monitoring of rice pests and diseases in a large area by multi-data fusion according to claim 1, characterized in that: The effectiveness test of the multi-scale data fusion sample set in S34 is as follows: S341. Calculate the statistical characteristics of the multi-scale data fusion sample set, including the mean value, standard deviation, and confidence interval, check the outliers, and remove the sample values ​​that are blocked by clouds or fog or interfered by mixed ground objects; S342. Calculate the correlation between the occurrence index of pests and diseases in satellite pixels and the occurrence of ground sampling points in the corresponding pixels. The Pearson correlation coefficient exceeds 0.5, indicating that the calculation method of the occurrence of rice pests and diseases in the multi-scale data fusion sample set is reasonable. S343. Count the spatial positions of the centers of all sampled satellite pixels and calculate the nearest neighbor distances to other center points. Then use the KS test to determine the difference between the empirical distribution function and the theoretical distribution function of the distance. If there is a significant difference, it means that the sampling positions of the satellite pixels obey the random distribution assumption.

6. The method for remote sensing monitoring of rice pests and diseases in a large area by multi-data fusion according to claim 1, characterized in that: In S4, the five-fold cross-validation method is used to divide the multi-scale data fusion sample set into multiple subsets. The base model is a machine learning algorithm, including logistic regression, decision tree, gradient boosting tree, support vector machine, KNN, and naive Bayes; the meta-model is a random forest model, which can learn how to combine the prediction results of the base model to maximize the accuracy of the overall model.

7. A large-area rice pest and disease remote sensing monitoring system with multi-data fusion, characterized by: A method for remote sensing monitoring of rice pests and diseases in a large area using multi-data fusion applied to any one of claims 1 to 6, comprising a ground sampling module, carrying a mobile verification APP and an Internet of Things monitoring device, wherein the mobile verification APP is used for sampling and storing ground data, and the Internet of Things monitoring device is used for continuously monitoring the distribution quantity and population change dynamics of field pests within the surrounding area; The UAV monitoring module is used to achieve sample category balance processing and extract rice pest and disease areas on the UAV image and transmit them to the data fusion module; The data fusion module is used to fuse ground data, UAV image data, and satellite image data at multiple scales to form a multi-scale data fusion sample set, and to test the effectiveness of the multi-scale data fusion sample set; The satellite monitoring module is used to perform integrated modeling training of large-area rice disease and insect pest identification models based on the constructed multi-scale data fusion sample set, and to call the trained large-area rice disease and insect pest identification model to predict satellite image data of large areas, and to obtain large-area rice disease and insect pest remote sensing monitoring results based on the rice disease and insect pest occurrence degree index of different pixels in the prediction results.

8. The multi-data fusion large-area rice pest remote sensing monitoring system according to claim 7 is characterized in that: The mobile verification APP consists of mobile device terminals, data transmission layer, background management system, user service layer and GIS service system: Mobile device terminals include smartphones and tablets, which are used as carriers to install mobile verification APP to collect on-site ground data; Data transmission layer: Output the on-site ground data obtained by the mobile phone or tablet camera, and the longitude and latitude data obtained by the positioning navigation system to the background management system through the mobile network; Backend management system: responsible for receiving, processing, storing and analyzing data transmitted from the data transmission layer; User service layer: set different permissions for users of different levels and open different data; GIS service system: responsible for the management and loading of different layers, as well as the display and release of sampling points; The IoT monitoring equipment consists of an insect trap, an insect collector, a high-definition camera, a power supply component, a data transmission device, and an AI intelligent recognition system: Insect trap: uses light, electricity, and chemical insect traps to attract, kill, and disperse rice field pests; Pest collector: responsible for collecting the dead insect carcasses and spreading them on a display whiteboard for subsequent intelligent identification and counting, and for regular cleaning and removal of the insect carcasses; HD camera: regularly captures the distribution of pests on the whiteboard of the insect collector and transmits the HD pictures to the AI ​​intelligent recognition system; Power supply components: obtain energy input through solar panels to reduce equipment maintenance costs; Data transmission equipment: transmits the data collected by IoT devices to the data center. The communication methods include wireless communication and wired communication. AI intelligent identification system: Based on the AI ​​identification algorithm for pests and diseases, it automatically identifies and counts the high-definition pictures taken, and counts the types and quantity changes of pests.

Images