Wine grape disease space-time early warning method and device based on meteorological and remote sensing data and medium

Abstract

The invention discloses a wine grape disease space-time early warning method and device based on meteorological and remote sensing data and a medium, and relates to the technical field of agricultural disease control. The method comprises the steps of obtaining multi-point meteorological data of a multi-period wine grape planting area, and performing preprocessing to obtain multi-period meteorological data; a wine grape downy mildew space-time prediction model is constructed, and the wine grape downy mildew space-time prediction model responds to input multi-period meteorological data and remote sensing images and outputs a prediction result; performing spatial interpolation on a prediction result; performing geographical registration on the prediction result; and importing a prediction result after spatial interpolation and geographic registration processing into an ArcGIS system, and displaying a first region with a disease risk value higher than a set threshold value and a second region with a disease risk value lower than the set threshold value in a distinguishing manner through a thermodynamic diagram visualization mode. According to the invention, multi-temporal satellite images are used, meteorological data are combined, space and time sequence characteristics of diseases are considered, and accurate early warning of large-range downy mildew diseases is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural disease control, and more specifically, to a spatio-temporal early warning method, device and medium for diseases of wine grapes based on meteorological and remote sensing data. Background Art

[0002] Grape downy mildew (Plasmopara viticola) is a fungal disease caused by downy mildew, and is one of the most common and serious diseases in global grape cultivation. The occurrence and spread of downy mildew are closely related to climatic conditions, especially the changes in temperature and humidity. Therefore, early and accurate warning is crucial for disease management in grape cultivation. However, there are many limitations in the current monitoring and warning methods for downy mildew, and they are still unable to provide efficient and real-time support for disease prevention and control.

[0003] In the prior art, the monitoring methods for grape downy mildew mainly include the following categories: 1) Manual monitoring method: Traditional downy mildew monitoring relies on manual inspection and sampling analysis. Usually, professionals need to go to the vineyard to check whether the leaves of the plants are invaded by diseases. This method is not only time-consuming and laborious, but also easily affected by human factors, and the monitoring results have large errors. At the same time, manual monitoring cannot reflect the change trend of diseases in real time, resulting in the failure to give timely warnings about the occurrence and spread of diseases. 2) Prediction models based on meteorological data: Some studies have tried to establish prediction models for downy mildew based on meteorological data (such as temperature, humidity, precipitation, etc.). Common methods include performing correlation analysis on meteorological conditions and historical disease data and combining empirical formulas to predict the occurrence of diseases. However, these models often rely too much on simplified meteorological parameters, ignoring the complex spatio-temporal dynamic characteristics of disease occurrence, resulting in low prediction accuracy and lack of consideration of multi-dimensional factors in the disease development process (such as climate differences in different regions and non-linear characteristics of disease transmission). 3) Disease identification and early warning systems based on machine learning: With the development of artificial intelligence and machine learning technologies, in recent years, researchers have tried to use machine learning algorithms (such as support vector machines, decision trees, random forests, etc.) to build early warning models for grape downy mildew. By collecting various environmental information such as meteorological data and soil humidity in the vineyard, combined with sensor data and image analysis, disease prediction is carried out. These methods can provide higher prediction accuracy to a certain extent, but there are still the following problems: 1) Data diversity problem: Existing models usually only use a single data source (such as meteorological data or image data), unable to fully consider the spatio-temporal dynamic characteristics of disease transmission, resulting in weak generalization ability of the models. 2) Ignoring spatio-temporal characteristics: Many existing machine learning methods (such as traditional 2D convolutional neural networks) can only process static data and cannot effectively capture the spatio-temporal dependence relationship in the process of downy mildew transmission. 3) Insufficient handling of missing data: Most methods lack effective data completion mechanisms when facing data missing or sensor failures, resulting in a decrease in the accuracy of the early warning system. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a spatio-temporal early warning method, device and medium for diseases of wine-making grapes based on meteorological and remote sensing data, and uses a deep learning method combining spatio-temporal information, meteorological data and satellite multi-spectral data to accurately warn about diseases of wine-making grape downy mildew, so as to improve the disease management effect of grape downy mildew.

[0005] In the first aspect, the present invention provides a spatio-temporal early warning method for diseases of wine-making grapes based on meteorological and remote sensing data, and the method includes:

[0006] Obtain multi-period multi-point meteorological data of the wine-making grape planting area, and perform preprocessing to obtain multi-period meteorological data;

[0007] Construct a spatio-temporal prediction model for grapevine downy mildew. The spatio-temporal prediction model for grapevine downy mildew responds to the input multi-period meteorological data and remote sensing images and outputs a prediction result;

[0008] Perform spatial interpolation on the prediction result;

[0009] Perform georegistration on the prediction result;

[0010] Import the prediction result after spatial interpolation and georegistration processing into the ArcGIS system, and distinguish and display the first area with a disease risk value higher than the set threshold and the second area with a value lower than the set threshold through the heat map visualization method.

[0011] Furthermore, obtain multi-point meteorological data in the grapevine planting area in multiple periods and perform preprocessing to obtain multi-period meteorological data, including:

[0012] Obtain multi-point meteorological data at multiple locations in the grapevine planting area in multiple periods;

[0013] Perform data completion on the multi-point meteorological data at multiple locations in the grapevine planting area in multiple periods;

[0014] Obtain satellite remote sensing images of the grapevine planting area in multiple periods;

[0015] Perform operations of radiometric correction, geometric correction, image denoising and image cropping on the satellite remote sensing images of the grapevine planting area in multiple periods to obtain remote sensing images;

[0016] Perform spatial registration on the meteorological data and the remote sensing images, and map the meteorological data into the spatial coordinate system of the remote sensing images;

[0017] Perform temporal registration on the meteorological data and the remote sensing images. Based on linear interpolation, align the multi-time meteorological data and the multi-period remote sensing images in time to obtain multi-period meteorological data.

[0018] Furthermore, the multi-point meteorological data in the grapevine planting area in multiple periods includes daily average temperature, humidity, precipitation and wind speed.

[0019] Furthermore, the method for performing data completion on the multi-point meteorological data at multiple locations in the grapevine planting area in multiple periods includes using an LSTM time series prediction model to predict and complete through the correlation of front and back data to obtain meteorological data.

[0020] Furthermore, the spatio-temporal prediction model for grapevine downy mildew includes an input layer, a first processing layer, a second processing layer, a third processing layer, a fourth processing layer, a feature extraction layer and an output layer connected in sequence.

[0021] Further, the input layer is used to obtain multi-temporal three-dimensional data, which includes multi-temporal meteorological data and remote sensing images, and its size is (T, D, H, W); where T represents the number of time periods, D represents the depth of meteorological data and remote sensing image data, H represents the height of meteorological data and remote sensing image data, and W represents the width of meteorological data and remote sensing image data;

[0022] The first processing layer is used to: use the first convolutional layer to convert the data input by the input layer into a feature map with 64 channels, with an output size of (64, D / 2, H / 4, W / 4), and perform normalization through LayerNorm; where the first convolutional layer is expressed as Conv3d(T, 64, kernel_size=(3, 4, 4), stride=(2, 4, 4), padding=(1, 0, 0)), Conv3d represents a 3D convolutional layer, kernel_size represents the convolutional kernel size, stride represents the stride, and padding represents the edge data padding size;

[0023] The second processing layer is used to: perform a convolution operation on the output of the first processing layer using the second convolutional layer, expand the number of channels to 128, with an output size of (128, D / 2, H / 8, W / 8), and perform normalization through LayerNorm; where the second convolutional layer is expressed as Conv3d(64, 128, kernel_size=(1, 2, 2), stride=(1, 2, 2));

[0024] The third processing layer is used to: in response to the output of the second processing layer, further increase the number of channels to 320 through the third convolutional layer, with an output size of (320, D / 2, H / 16, W / 16), and perform LayerNorm normalization; where the third convolutional layer is expressed as Conv3d(128, 320, kernel_size=(1, 2, 2), stride=(1, 2, 2));

[0025] The fourth processing layer is used to: in response to the output of the third processing layer, perform a convolution operation using the fourth convolutional layer, expand the number of channels to 512, with an output size of (512, D / 2, H / 32, W / 32), and perform normalization through LayerNorm; where the fourth convolutional layer is expressed as Conv3d(320, 512, kernel_size=(1, 2, 2), stride=(1, 2, 2));

[0026] The feature extraction layer includes multiple convolutional blocks, each convolutional block includes Conv3d, BatchNorm3d and positional encoding, and is used to further extract and process the input features; wherein, the input features are the output of the fourth processing layer.

[0027] The output layer is used to output the prediction result of the corresponding area according to the processing result of the feature extraction layer.

[0028] Further, performing spatial interpolation on the prediction result includes:

[0029] Using the methods of bilinear interpolation and cubic interpolation to perform spatial interpolation on the prediction result to effectively smooth the data; wherein, for the prediction results of high-dimensional and complex regions, Kriging interpolation is used to improve the interpolation accuracy.

[0030] Further, performing georegistration on the prediction result includes:

[0031] Based on the precise coordinates of the prediction points and the precise coordinates of the corresponding remote sensing data, by inputting the known coordinate values, calculate the registration transformation matrix, and map the remote sensing image and meteorological data into a unified coordinate system to complete the registration.

[0032] After completing the registration, compare the registration result with the actual observation data to determine the registration error. If the registration error exceeds the set threshold range, then use spatial resampling and error correction for adjustment to make the registration error within the set threshold range.

[0033] In a second aspect, the present invention provides a spatio-temporal early warning device for grapevine diseases based on meteorological and remote sensing data, and the device includes:

[0034] A data preprocessing unit, configured to obtain multi-period multi-point meteorological data of the grapevine planting area and perform preprocessing to obtain multi-period meteorological data.

[0035] A model prediction unit, configured to construct a spatio-temporal prediction model for grapevine downy mildew, and the spatio-temporal prediction model for grapevine downy mildew outputs a prediction result in response to the input multi-period meteorological data and remote sensing images.

[0036] A spatial interpolation unit, configured to perform spatial interpolation on the prediction result.

[0037] A georegistration unit, configured to perform georegistration on the prediction result.

[0038] A result display unit, configured to import the prediction result after spatial interpolation and georegistration processing into the ArcGIS system, and distinguish and display the first area where the disease risk value is higher than the set threshold and the second area where the disease risk value is lower than the set threshold through the heat map visualization method.

[0039] In a third aspect, the present invention provides a readable storage medium storing one or more programs, which can be executed by one or more processors to implement the method as described above.

[0040] The present invention has at least the following beneficial effects:

[0041] The present invention uses multi-temporal satellite images, combined with meteorological data, taking into account the spatial and temporal characteristics of the disease, and realizing accurate early warning of downy mildew disease in a large area. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Shows an overall flowchart of a spatio-temporal early warning method for grapevine diseases based on meteorological and remote sensing data according to an embodiment of the present invention.

[0043] Figure 2 Shows a flowchart for preprocessing multi-point meteorological data in a multi-period grapevine planting area according to an embodiment of the present invention;

[0044] Figure 3 Shows a structural diagram of a spatio-temporal prediction model for grapevine downy mildew according to an embodiment of the present invention;

[0045] Figure 4 Shows a structural diagram of a spatio-temporal early warning device for grapevine diseases based on meteorological and remote sensing data according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings and specific examples, but shall not be construed as a limitation to the present invention. For the various steps described herein, if there is no necessity for a sequential relationship between them, the order in which they are described as examples herein shall not be construed as a limitation, and those skilled in the art should know that they can be adjusted in order as long as the logic between them is not destroyed and the entire process cannot be realized.

[0047] An embodiment of the present invention provides a spatio-temporal early warning method for grapevine diseases based on meteorological and remote sensing data, as Figure 1 shown, the spatio-temporal early warning method for grapevine diseases based on meteorological and remote sensing data includes the following steps S10 to S50.

[0048] S10. Obtain multi-point meteorological data in a multi-period grapevine planting area, and perform preprocessing to obtain multi-period meteorological data.

[0049] In some embodiments, as Figure 1As shown, step S10 includes the following steps S11 to S16.

[0050] S11, obtain multi - period meteorological data at multiple points in the wine grape planting area.

[0051] In this embodiment, obtaining multi - period meteorological data at multiple points in the wine grape planting area includes daily average temperature, humidity, precipitation, and wind speed.

[0052] S12, perform data completion on the multi - period meteorological data at multiple points in the wine grape planting area.

[0053] In this embodiment, step S12 performs data completion on the multi - period meteorological data obtained in step S11, using an LSTM time - series prediction model to predict and complete through the correlation of front - and - back data.

[0054] S13, obtain satellite remote - sensing images of the multi - period wine grape planting area.

[0055] S14, perform operations of radiometric correction, geometric correction, image denoising, and image cropping on the satellite remote - sensing images of the multi - period wine grape planting area to obtain remote - sensing images.

[0056] S15, perform spatial registration on the meteorological data and the remote - sensing images, and map the meteorological data into the spatial coordinate system of the remote - sensing images.

[0057] Exemplarily, when performing spatial registration on the meteorological data and the remote - sensing images, use Geographic Information System (GIS) technology to map the multi - period meteorological data obtained in step S12 into the spatial coordinate system of the remote - sensing images obtained in step S14.

[0058] S16, perform temporal registration on the meteorological data and the remote - sensing images. Based on linear interpolation, align the multi - time meteorological data and the multi - period remote - sensing images in time to obtain multi - period meteorological data.

[0059] S20, construct a spatio - temporal prediction model for grapevine downy mildew. The spatio - temporal prediction model for grapevine downy mildew responds to the input multi - period meteorological data and remote - sensing images and outputs a prediction result.

[0060] In some embodiments, as Figure 3 shown, is the structural diagram of the spatio - temporal prediction model for grapevine downy mildew. The spatio - temporal prediction model for grapevine downy mildew includes an input layer Input, a first processing layer PatchEmbed1, a second processing layer PatchEmbed2, a third processing layer PatchEmbed3, a fourth processing layer PatchEmbed4, a feature extraction layer CBlock, and an output layer Output that are connected in sequence.

[0061] Specifically, the structures and functions of each layer are as follows:

[0062] The input layer Input is used to obtain multi - period 3D data, which includes multi - period meteorological data and remote sensing images, and its size is (T, D, H, W); where T represents the number of periods, D represents the depth of meteorological data and remote sensing image data, H represents the height of meteorological data and remote sensing image data, and W represents the width of meteorological data and remote sensing image data;

[0063] The first processing layer PatchEmbed1 is used to: convert the data input by the input layer into a feature map with 64 channels using the first convolutional layer, with an output size of (64, D / 2, H / 4, W / 4), and perform normalization through LayerNorm; where the first convolutional layer is represented as Conv3d(T, 64, kernel_size=(3, 4, 4), stride=(2, 4, 4), padding=(1, 0, 0)), Conv3d represents a 3D convolutional layer, kernel_size represents the convolutional kernel size, stride represents the stride, and padding represents the edge data padding size;

[0064] The second processing layer PatchEmbed2 is used to: perform a convolutional operation on the output of the first processing layer using the second convolutional layer, expand the number of channels to 128, with an output size of (128, D / 2, H / 8, W / 8), and perform normalization through LayerNorm; where the second convolutional layer is represented as Conv3d(64, 128, kernel_size=(1, 2, 2), stride=(1, 2, 2));

[0065] The third processing layer PatchEmbed3 is used to: in response to the output of the second processing layer, further increase the number of channels to 320 through the third convolutional layer, with an output size of (320, D / 2, H / 16, W / 16), and perform LayerNorm normalization; where the third convolutional layer is represented as Conv3d(128, 320, kernel_size=(1, 2, 2), stride=(1, 2, 2));

[0066] The fourth processing layer PatchEmbed4 is used to: in response to the output of the third processing layer, perform a convolutional operation using the fourth convolutional layer, expand the number of channels to 512, with an output size of (512, D / 2, H / 32, W / 32), and perform normalization through LayerNorm; where the fourth convolutional layer is represented as Conv3d(320, 512, kernel_size=(1, 2, 2), stride=(1, 2, 2));

[0067] The feature extraction layer CBlock includes multiple convolutional blocks, each of which includes Conv3d, BatchNorm3d, and positional encoding, and is used to further extract and process the input features; wherein, the input features are the output of the fourth processing layer.

[0068] The output layer Output is used to output the prediction results of the corresponding area according to the processing results of the feature extraction layer.

[0069] S30, perform spatial interpolation on the prediction results.

[0070] In some embodiments, spatial interpolation technology is used to resample the prediction results. The methods of bilinear interpolation and cubic interpolation are used to effectively smooth the data, and while maintaining the edge information, ensure that the interpolation results have high accuracy. For high-dimensional and complex regions, Kriging interpolation is used to further improve the interpolation accuracy. Spatial interpolation can adjust the prediction results of different resolutions to the resolution of the remote sensing image, so as to ensure that the prediction results are spatially aligned with the image data and reduce the errors caused by resolution differences.

[0071] S40, perform georegistration on the prediction results.

[0072] In some embodiments, select the precise coordinates of a prediction point and the precise coordinates of the corresponding remote sensing data, use the spatial registration tool of the ARCGIS geographic information system, calculate the registration transformation matrix by inputting the known coordinate values, and map the remote sensing image and meteorological data into a unified coordinate system (WGS 84). After completing the registration, compare the registration results with the actual observation data (such as prediction result data or landmark points with known geographical locations) to further verify the accuracy of the registration. If there is a registration error, use spatial resampling and error correction techniques to adjust to ensure the high accuracy of the final registration result.

[0073] S50, import the prediction results after spatial interpolation and georegistration processing into the ArcGIS system, and distinguish and display the first area with a disease risk value higher than the set threshold and the second area with a value lower than the set threshold through the heatmap visualization method.

[0074] Exemplarily, when presenting the prediction results in step S50, import the spatio-temporal prediction results after spatial interpolation and georegistration processing into the ArcGIS system. The areas with higher disease risk values are represented in red by the heatmap visualization method, while the low-risk areas are represented in green.

[0075] The embodiment of the present invention also provides a spatio-temporal early warning device for grapevine diseases based on meteorological and remote sensing data, asFigure 4 As shown in Figure 4 , the device includes:

[0076] A data preprocessing unit 401, configured to obtain multi - period multi - point meteorological data of the wine grape planting area, and perform preprocessing to obtain multi - period meteorological data;

[0077] A model prediction unit 402, configured to construct a spatio - temporal prediction model for grapevine downy mildew. The spatio - temporal prediction model for grapevine downy mildew outputs a prediction result in response to the input multi - period meteorological data and remote sensing images;

[0078] A spatial interpolation unit 403, configured to perform spatial interpolation on the prediction result;

[0079] A georegistration unit 404, configured to perform georegistration on the prediction result;

[0080] A result display unit 405, configured to import the prediction result after spatial interpolation and georegistration processing into the ArcGIS system, and distinguish and display the first area where the disease risk value is higher than the set threshold and the second area where the disease risk value is lower than the set threshold through the heat map visualization method.

[0081] In some embodiments, the data preprocessing unit is further configured to:

[0082] Obtain multi - period multi - point meteorological data of the wine grape planting area;

[0083] Perform data completion on the multi - period multi - point meteorological data of the wine grape planting area;

[0084] Obtain satellite remote sensing images of the wine grape planting area in multiple periods;

[0085] Perform operations of radiometric correction, geometric correction, image denoising, and image cropping on the satellite remote sensing images of the wine grape planting area in multiple periods to obtain remote sensing images;

[0086] Perform spatial registration on the meteorological data and the remote sensing images, and map the meteorological data into the spatial coordinate system of the remote sensing images;

[0087] Perform temporal registration on the meteorological data and the remote sensing images. Based on linear interpolation, align the multi - time meteorological data and the multi - period remote sensing images in time to obtain multi - period meteorological data.

[0088] In some embodiments, the multi - period multi - point meteorological data of the wine grape planting area includes daily average temperature, humidity, precipitation, and wind speed.

[0089] In some embodiments, the method for completing the multi - point meteorological data of the multi - period wine grape planting area includes using an LSTM time - series prediction model to predict and complete the data through the correlation of the front and back data, so as to obtain the meteorological data.

[0090] In some embodiments, the spatio - temporal prediction model for grapevine downy mildew includes an input layer, a first processing layer, a second processing layer, a third processing layer, a fourth processing layer, a feature extraction layer, and an output layer connected in sequence.

[0091] In some embodiments, the input layer is used to obtain multi - period three - dimensional data, and the multi - period three - dimensional data includes multi - period meteorological data and remote sensing images, and its size is (T, D, H, W); where T represents the number of periods, D represents the depth of the meteorological data and remote sensing image data, H represents the height of the meteorological data and remote sensing image data, and W represents the width of the meteorological data and remote sensing image data.

[0092] The first processing layer is used to: use the first convolutional layer to convert the data input by the input layer into a feature map with 64 channels, and the output size is (64, D / 2, H / 4, W / 4), and perform normalization through LayerNorm; where the first convolutional layer is represented as Conv3d(T, 64, kernel_size=(3, 4, 4), stride=(2, 4, 4), padding=(1, 0, 0)), Conv3d represents a 3D convolutional layer, kernel_size represents the convolutional kernel size, stride represents the stride, and padding represents the edge data padding size.

[0093] The second processing layer is used to: perform a convolutional operation on the output of the first processing layer using the second convolutional layer to expand the number of channels to 128, and the output size is (128, D / 2, H / 8, W / 8), and perform normalization through LayerNorm; where the second convolutional layer is represented as Conv3d(64, 128, kernel_size=(1, 2, 2), stride=(1, 2, 2)).

[0094] The third processing layer is used to: in response to the output of the second processing layer, further increase the number of channels to 320 through the third convolutional layer, and the output size is (320, D / 2, H / 16, W / 16), and perform LayerNorm normalization; where the third convolutional layer is represented as Conv3d(128, 320, kernel_size=(1, 2, 2), stride=(1, 2, 2)).

[0095] The fourth processing layer is used to: in response to the output of the third processing layer, perform a convolution operation using a fourth convolutional layer to expand the number of channels to 512, with an output size of (512, D / 2, H / 32, W / 32), and perform normalization through LayerNorm; wherein, the fourth convolutional layer is represented as Conv3d(320, 512, kernel_size=(1, 2, 2), stride=(1, 2, 2)).

[0096] The feature extraction layer includes a plurality of convolutional blocks, each convolutional block including Conv3d, BatchNorm3d, and positional encoding, for further extracting and processing the input features; wherein, the input features are the output of the fourth processing layer.

[0097] The output layer is used to output the prediction result of the corresponding area according to the processing result of the feature extraction layer.

[0098] In some embodiments, the spatial interpolation unit is further configured to:

[0099] Perform spatial interpolation on the prediction result by using the methods of bilinear interpolation and cubic interpolation to effectively smooth the data; wherein, for the prediction results of high-dimensional and complex regions, Kriging interpolation is used to improve the interpolation accuracy.

[0100] In some embodiments, the georegistration unit is further configured to:

[0101] Based on the precise coordinates of the prediction points and the precise coordinates of the corresponding remote sensing data, calculate the registration transformation matrix through the input known coordinate values, and map the remote sensing image and meteorological data into a unified coordinate system to complete the registration;

[0102] After completing the registration, compare the registration result with the actual observation data to determine the registration error. If the registration error exceeds the set threshold range, then perform adjustment by using spatial resampling and error correction so that the registration error is within the set threshold range.

[0103] It should be noted that the structures of the various wine grape disease spatio-temporal warning devices based on meteorological and remote sensing data described in this embodiment belong to the same technical concept as the previously described wine grape disease spatio-temporal warning method based on meteorological and remote sensing data, and achieve the same beneficial effects through the same principle, which will not be elaborated here.

[0104] The embodiment of the present invention also provides a readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the method described in any one of the above embodiments.

[0105] The foregoing description is intended to be illustrative and not restrictive. For example, the above examples (or one or more aspects thereof) may be used in combination with each other. For instance, those of ordinary skill in the art may use other embodiments when reading the above description. Additionally, in the above detailed description, various features may be grouped together to simplify the present invention. This should not be construed as an intention that the features of an unclaimed invention are necessary for any claim. On the contrary, the subject matter of the present invention may be less than all of the features of a particular embodiment of the invention. Thus, the following claims are hereby incorporated into the detailed description by way of example or embodiment, where each claim stands on its own as a separate embodiment, and it is contemplated that these embodiments may be combined with each other in various combinations or permutations. The scope of the present invention should be determined with reference to the appended claims and the full scope of equivalents to which those claims are entitled.

Claims

1. A spatiotemporal early warning method for wine grape diseases based on meteorological and remote sensing data, characterized in that: The method comprises: Acquire multi-point meteorological data of wine grape planting areas in multiple periods, and perform preprocessing to obtain multi-period meteorological data; Constructing a spatiotemporal prediction model for downy mildew of wine grapes, wherein the spatiotemporal prediction model for downy mildew of wine grapes responds to input multi-period meteorological data and remote sensing images and outputs prediction results; Performing spatial interpolation on the prediction results; georeferencing the prediction results; The prediction results after spatial interpolation and geo-referenced processing are imported into the ArcGIS system, and the first area with disease risk value higher than the set threshold and the second area with disease risk value lower than the set threshold are displayed through heat map visualization.

2. The spatiotemporal early warning method for wine grape diseases based on meteorological and remote sensing data according to claim 1, characterized in that: Acquire multi-point meteorological data of wine grape planting areas in multiple periods and perform preprocessing to obtain multi-period meteorological data, including: Obtain multi-point meteorological data for wine grape growing areas over multiple periods; Completing the multi-point meteorological data of the wine grape planting area in the multi-periods; Acquire satellite remote sensing images of wine grape growing areas over multiple periods; Performing radiation correction, ensemble correction, image denoising and image cropping operations on the satellite remote sensing images of the wine grape planting area in the multiple periods to obtain remote sensing images; spatially registering the meteorological data with the remote sensing image, mapping the meteorological data to the spatial coordinate system of the remote sensing image; The meteorological data and the remote sensing images are temporally registered, and based on linear interpolation, the multi-time meteorological data and the multi-period remote sensing images are temporally aligned to obtain the multi-period meteorological data.

3. The spatiotemporal early warning method for wine grape diseases based on meteorological and remote sensing data according to claim 1, characterized in that: The multi-point meteorological data of the wine grape growing area in multiple periods include average daily temperature, humidity, precipitation and wind speed.

4. The spatiotemporal early warning method for wine grape diseases based on meteorological and remote sensing data according to claim 2, characterized in that: The method of completing the multi-point meteorological data of the wine grape planting area in the multi-periods includes adopting an LSTM time series prediction model to predict and complete the meteorological data through the correlation between the previous and next data.

5. The spatiotemporal early warning method for wine grape diseases based on meteorological and remote sensing data according to claim 1, characterized in that: The spatiotemporal prediction model for downy mildew of wine grapes comprises an input layer, a first processing layer, a second processing layer, a third processing layer, a fourth processing layer, a feature extraction layer and an output layer which are connected in sequence.

6. The spatiotemporal early warning method for wine grape diseases based on meteorological and remote sensing data according to claim 5, characterized in that: The input layer is used to obtain multi-period three-dimensional data, which includes multi-period meteorological data and remote sensing images, and its size is (T, D, H, W); wherein T represents the number of periods, D represents the depth of meteorological data and remote sensing image data, H represents the height of meteorological data and remote sensing image data, and W represents the width of meteorological data and remote sensing image data; The first processing layer is used to: use the first convolution layer to convert the data input by the input layer into a 64-channel feature map, the output size is (64, D / 2, H / 4, W / 4), and normalize it through LayerNorm; wherein the first convolution layer is represented as Conv3d(T, 64, kernel_size = (3, 4, 4), stride = (2, 4, 4), padding = (1, 0, 0)), Conv3d represents a 3D convolution layer, kernel_size represents the convolution kernel size, stride represents the step size, and padding represents the edge data padding size; The second processing layer is used to: use the second convolutional layer to perform a convolution operation on the output of the first processing layer, expand the number of channels to 128, and the output size is (128, D / 2, H / 8, W / 8), and normalize it through LayerNorm; wherein the second convolutional layer is represented as Conv3d(64, 128, kernel_size = (1, 2, 2), stride = (1, 2, 2)); The third processing layer is used to: in response to the output of the second processing layer, further increase the number of channels to 320 through the third convolution layer, the output size is (320, D / 2, H / 16, W / 16), and perform LayerNorm normalization; wherein the third convolution layer is represented as Conv3d(128, 320, kernel_size = (1, 2, 2), stride = (1, 2, 2)); The fourth processing layer is used to: respond to the output of the third processing layer, use the fourth convolutional layer to perform a convolution operation, expand the number of channels to 512, and the output size is (512, D / 2, H / 32, W / 32), and normalize by LayerNorm; wherein the fourth convolutional layer is represented as Conv3d(320, 512, kernel_size = (1, 2, 2), stride = (1, 2, 2)); The feature extraction layer includes a plurality of convolution blocks, each of which includes Conv3d, BatchNorm3d and position encoding, for further extracting and processing input features; wherein the input features are outputs of the fourth processing layer; The output layer is used to output the prediction result of the corresponding area according to the processing result of the feature extraction layer.

7. The spatiotemporal early warning method for wine grape diseases based on meteorological and remote sensing data according to claim 1, characterized in that: The prediction result is spatially interpolated, including: The prediction results are spatially interpolated using bilinear interpolation and cubic interpolation methods to effectively smooth the data; for prediction results in high dimensions and complex areas, Kriging interpolation is used to improve interpolation accuracy.

8. The spatiotemporal early warning method for wine grape diseases based on meteorological and remote sensing data according to claim 1, characterized in that: The prediction results are georeferenced, including: Based on the precise coordinates of the predicted points and the precise coordinates of the corresponding remote sensing data, the registration transformation matrix is ​​calculated by inputting the known coordinate values, and the remote sensing images and meteorological data are mapped into a unified coordinate system to complete the registration; After the registration is completed, the registration result is compared with the actual observation data to determine the registration error. If the registration error exceeds the set threshold range, spatial resampling and error correction are used to adjust it so that the registration error is within the set threshold range.

9. A spatiotemporal early warning device for wine grape diseases based on meteorological and remote sensing data, characterized in that: The device comprises: A data preprocessing unit is configured to obtain multi-point meteorological data of wine grape planting areas in multiple periods, and perform preprocessing to obtain multi-period meteorological data; A model prediction unit is configured to construct a spatiotemporal prediction model for downy mildew of wine grapes, wherein the spatiotemporal prediction model for downy mildew of wine grapes responds to input multi-period meteorological data and remote sensing images and outputs prediction results; A spatial interpolation unit, configured to perform spatial interpolation on the prediction result; A geo-referencing unit, configured to perform geo-referencing on the prediction result; The result display unit is configured to import the prediction results after spatial interpolation and geographic registration into the ArcGIS system, and to distinguish between the first area where the disease risk value is higher than the set threshold and the second area where the disease risk value is lower than the set threshold through heat map visualization. 10 . A non-transitory computer-readable storage medium storing instructions, which, when executed by a processor, perform the method according to claim 1 .

Images