WO2024019632A1

WO2024019632A1 - Device and method for determining crop productivity

Info

Publication number: WO2024019632A1
Application number: PCT/RU2022/000234
Authority: WO
Inventors: Александр Леонидович АНОХИН; Борис Витальевич ЦХАЙ; Лев Владимирович ТЕМИН
Original assignee: Публичное Акционерное Общество "Сбербанк России"
Priority date: 2022-07-22
Filing date: 2022-07-22
Publication date: 2024-01-25

Abstract

The technical solution relates to the field of computing. The claimed method for determining the productivity of an agricultural field is computer implemented. The method includes receiving a request to determine the productivity of a field, and receiving a first set of multispectral images corresponding to a given timeframe, which contains values of VH and VV polarization and of the angle of incidence during synthetic aperture radar imaging, said first set being used to determine a set of polarization indices for the agricultural field. A second set of multispectral images corresponding to the given timeframe is received which contains intensity values of pixels in the visible infrared spectrum, the near-infrared spectrum and the shortwave infrared spectrum, as well as values of the cloudiness of the image pixels, said second set being used to determine a set of vegetation indices for the agricultural field and a set of colour indices. Soil moisture content indices for the agricultural field and also weather data are determined for the given timeframe. The productivity of the agricultural field is determined using data characterizing the crop type, the set of polarization and vegetation indices, the set of colour indices, the soil moisture content indices, and the weather data. This provides for more accurate prediction of the productivity of an agricultural field.

Description

DEVICE AND METHOD FOR DETERMINING THE YIELD OF AGRICULTURAL CROPS

TECHNICAL FIELD

[0001] The presented technical solution relates, in general, to the field of computer technology, and in particular to a method and device for determining the yield of agricultural crops using deep learning based on Earth remote sensing (ERS) data and weather data, in including the Sentinel-1, Sentinel-2 and MetOp ASCAT constellations. Sentinel-1 and 2, MetOp ASCAT - a family of satellites of the European Space Agency, created as part of the Copernicus global environmental monitoring and security project, designed to monitor the use of land, vegetation, forest and water resources, can also be used in disaster response .

BACKGROUND OF THE ART

[0002] Currently, technologies for monitoring the condition of agricultural fields using remote sensing data have become widespread.

[0003] The most similar solution is the method for determining the yield of agricultural crops, disclosed in patent US10769733B2, publ. 09/08/2020. Having similar approaches, this method uses predominantly weather data, while the presented solution is based not only on feature descriptions such as temperature, humidity and precipitation, but also on data obtained from images, for example, the Sentinel-1 and Sentinel-1 constellations. 2, and also uses vegetation indices, humidity data, weather observations and crop rotation history with actual yields for specific fields. This approach allows us to create a more complete representation of the feature description.

[0004] US10769733B2 proposed to use a linear regression model as a model. The presented solution is based on the gradient growing tree model, which is a more powerful model compared to linear regression. In addition, for each agricultural crop, its own model is trained with its own parameters to achieve maximum prediction quality. [0005] A related solution is the system and method for predicting yield, disclosed in application US20200380617A1, publ. 03.12.2021. This solution uses a similar feature description, but there is a difference in the model itself. To predict crop yields, it was proposed to use multiple linear regression, however, as the authors of the application themselves note, this solution did not provide the best quality. The presented solution is based on the model of gradiently growing decision trees. Such a model allows not only to better identify patterns and show higher accuracy, but is also quite inexpensive. In the presented solution, the feature description of the image is encoded using autoencoders, which makes it possible to significantly improve the quality of prediction compared to using the original data.

[0006] Also a similar solution is a method for predicting crop yields during the growing season, disclosed in application US20200183048A1, publ. 06/11/2020. The difference between the presented solution lies in the processing of image features. This method uses linear regression to predict crop yields, which is a weaker model compared to the presented solution, where each model is selected individually for each crop.

DISCLOSURE OF INVENTION

[0007] The technical problem or task posed in this technical solution is to create a new, effective, simple and reliable method for determining the yield of an agricultural (agricultural) field.

[0008] The technical result is to increase the accuracy of forecasting the yield of an agricultural field.

[0009] The specified technical result is achieved by implementing a method for determining the yield of an agricultural field, performed by at least one computing device, containing the stages of:

- receive a request to determine the yield of the field, containing data characterizing the type of crop;

- obtain the first set of multispectral images for a given period of time, and each multispectral image of an agricultural field contains wave polarization values, in particular the values of VH polarization, W polarization and the angle of incidence of the beam when shooting by synthetic aperture radar; - based on the mentioned first set of images, a set of polarization indices for the agricultural field is determined;

- obtain a second set of multispectral images for a given period of time, and each multispectral image of an agricultural field contains pixel intensity values in the visible infrared spectrum, near-infrared spectrum and short-wave infrared spectrum, as well as cloudiness values of image pixels;

- based on the mentioned second set of images, a set of vegetation indices for an agricultural field and a set of color indicators are determined;

- determine soil moisture content indices for a given period of time for an agricultural field;

- determine weather data for a given period of time for an agricultural field;

- based on data characterizing the type of crop, a set of polarization indices, a set of vegetation indices, a set of color indicators, soil moisture content indices and weather data, the yield of an agricultural field is determined.

[0010] In one particular example of the method, the yield of an agricultural field is determined taking into account data characterizing the history of crop rotation and the actual yield for a given period of time.

[0011] In another particular example of the method, soil moisture content indices for a given period of time for an agricultural field are determined based on data from a third set of multispectral images for a given period of time obtained using at least one satellite designed to monitor soil moisture content.

[0012] In another particular example of the method, the stage of determining a set of polarization indices for an agricultural field contains stages in which:

- extract from the first set of multispectral images for each pixel of the agricultural field the values of VH polarization, W polarization and angle of incidence of the radar aperture synthesis beam;

- determined for each pixel of the agricultural field based on the extracted values: the value of the standard deviation of VH polarization; polarization standard deviation value W; the value of the standard deviation of the angle of incidence of the beam when shooting using the synthetic aperture radar method;

- determine the average values of VH polarization, W polarization, angle of incidence of the beam and the deviation values defined above; - assign average values of VH polarization, W polarization, angle of incidence of the beam and the deviation values defined above as polarization indices to the agricultural field for each multispectral image of the agricultural field in the first set;

- convert the values of the set of polarization indices obtained at the previous stage using the encoder model into a fixed vector.

[0013] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range (664.6 nm) are extracted and on their basis the value of the normalized vegetation index is determined for each pixel (NDVI);

- based on the NDVI values of the pixels, the average NDVI value is determined, which is assigned as the NDVI index to the agricultural field.

[0014] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the values of the difference vegetation index (DVI) are determined for each pixel x fields;

- based on the DVI values of the pixels, the average DVI value is determined, which is assigned as the DVI index to the agricultural field.

[0015] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR), the pixel intensity value in the red range and the pixel intensity value in the blue range (492.4 nm.) are extracted, and then on their on a basis for each pixel the value of the improved vegetation index (EVI) is determined;

- based on the EVI values of the pixels, the average EVI value is determined, which is assigned as the EVI index to the agricultural field;

- determine the leaf surface index based on the EVI index value

(LAI). [0016] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted, after which, on their basis, the value of the global environmental monitoring index is determined for each pixel ( Gemi);

- based on the Gemi values of the pixels, the average Gemi value is determined, which is assigned as the Gemi index to the agricultural field.

[0017] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR), the pixel intensity value in the blue range, the pixel intensity value in the green range (559.8 nm.) and the pixel intensity value in in the red range and on their basis for each pixel the value of the green vegetation index resistant to atmospheric influences (Gari) is determined;

- based on the Gari values of the pixels, the average Gari value is determined, which is assigned as the Gari index to the agricultural field.

[0018] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the green range are extracted and on their basis, the green difference vegetation index (GDVI) value is determined for each pixel ;

- based on the GDVI values of the pixels, the average GDVI value is determined, which is assigned as the Gari index to the agricultural field.

[0019] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the blue range, the pixel intensity value in the green range and the value pixel intensities in the red range and on their basis the value of the leaf greenness index (GLI) is determined for each pixel;

- based on the GLI values of the pixels, the average GLI value is determined, which is assigned as the GLI index to the agricultural field.

[0020] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the green range are extracted and on their basis the value of the optimized soil vegetation index (GOSAVI) is determined for each pixel;

- based on the GOSAVI values of the pixels, the average GOSAVI value is determined, which is assigned as the GOSAVI index to the agricultural field.

[0021] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the green range are extracted and on their basis the value of the soil greenness index (GSAVI) is determined for each pixel;

- based on the GSAVI values of the pixels, the average GSAVI value is determined, which is assigned as the GSAVI index to the agricultural field.

[0022] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the infrared vegetation index (IPVI) value is determined for each pixel;

- based on the IPVI values of the pixels, the average IPVI value is determined, which is assigned as the IPVI index to the agricultural field.

[0023] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near field is extracted infrared spectrum (NIR) and the pixel intensity value in the red range and on their basis for each pixel the value of the modified nonlinear vegetation index (MNLI) is determined;

- based on the MNLI values of the pixels, the average MNLI value is determined, which is assigned as the MNLI index to the agricultural field.

[0024] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the value of the pixel intensity in the near-infrared spectrum (NIR) and the value of the pixel intensity in the red range are extracted and on their basis for each pixel the value of the modified vegetation index with soil correction is determined ( MSAVI2);

- based on the MSAVI2 pixel values, the average MSAVI2 value is determined, which is assigned as the MSAVI2 index to the agricultural field.

[0025] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the value of the nonlinear vegetation index (NLI) is determined for each pixel;

- based on the NLI values of the pixels, the average NLI value is determined, which is assigned as the NLI index to the agricultural field.

[0026] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis for each pixel the value of the soil-optimized vegetation index (OSAVI) is determined );

- based on the OSAVI values of the pixels, the average OSAVI value is determined, which is assigned as the OSAVI index to the agricultural field.

[0027] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which: - from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the value of the renormalized difference vegetation index (RDVI) is determined for each pixel;

- based on the RDVI values of the pixels, the average RDVI value is determined, which is assigned as the RDVI index to the agricultural field.

[0028] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the soil vegetation index (SAVI) value is determined for each pixel;

- based on the SAVI values of the pixels, the average SAVI value is determined, which is assigned as the SAVI index to the agricultural field.

[0029] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the value of the transformed difference vegetation index (TDVI) is determined for each pixel;

- based on the TDVI values of the pixels, the average TDVI value is determined, which is assigned as the TDVI index to the agricultural field.

[0030] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the blue range, the pixel intensity value in the green range and the pixel intensity value in the red range are extracted and on their basis the value of the triangular vegetation index (TGI) is determined for each pixel );

- based on the TGI values of the pixels, the average TGI value is determined, which is assigned as the TGI index to the agricultural field. [0031] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the blue range, the pixel intensity value in the green range and the pixel intensity value in the red range are extracted and on their basis the value of the visible weatherproof vegetation index is determined for each pixel ( VARI);

- based on the VARI pixel values, the average VARI value is determined, which is assigned as the VARI index to the agricultural field.

[0032] In another particular example of the method, the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the value of the normalized difference vegetation index (WDRVI) is determined for each pixel;

- based on the WDRVI values of the pixels, the average WDRVI value is determined, which is assigned as the WDRVI index to the agricultural field.

[0033] In another particular example of the method, the steps are additionally performed in which:

- cloud values of image pixels are extracted from the second set of multispectral images of the agricultural field;

- based on the number of cloudy pixels, the cloudiness index of the agricultural field is determined.

[0034] In another particular example of the method, a step is additionally performed in which the obtained values of the vegetation indices assigned to the agricultural field are converted by means of an encoder model into a fixed vector.

[0035] In another particular example of the method, the stage of determining a set of color indicators for an agricultural field contains stages in which: the second set of multispectral images for each pixel is extracted from the received data:

- pixel intensity value in the ultrablue range, obtained at a resolution of 60 meters with a central wavelength of 443 nanometers; - pixel intensity value in the blue range, obtained at a resolution of 10 meters with a central wavelength of 490 nanometers;

- pixel intensity value in the green range obtained at a resolution of 10 meters with a central wavelength of 560 nanometers;

- pixel intensity value in the red range, obtained at a resolution of 10 meters with a central wavelength of 665 nanometers;

- pixel intensity value in visible and near-infrared (NIR) light obtained at a resolution of 20 meters with a central wavelength of 705 nanometers;

- pixel intensity value in visible and near-infrared (NIR) light obtained at a resolution of 20 meters with a central wavelength of 740 nanometers;

- pixel intensity value in visible and near-infrared (NIR) light obtained at a resolution of 20 meters with a central wavelength of 783 nanometers;

- pixel intensity value in visible and near-infrared (NIR) light obtained at a resolution of 10 meters with a central wavelength of 842 nanometers;

- pixel intensity value in visible and near-infrared (NIR) light obtained at a resolution of 120 meters with a central wavelength of 865 nanometers;

- pixel intensity value in short-wave infrared light (SWIR), obtained at a resolution of 60 meters with a central wavelength of 940 nanometers;

- pixel intensity value in short-wave infrared light (SWIR), obtained at a resolution of 60 meters with a central wavelength of 1375 nanometers;

- pixel intensity value in short-wave infrared light (SWIR), obtained at a resolution of 20 meters with a central wavelength of 1610 nanometers;

- pixel intensity value in short-wave infrared light (SWIR), obtained at a resolution of 20 meters with a central wavelength of 2190 nanometers;

- cloudiness values of agricultural field image pixels; based on the above-mentioned color indicator values extracted for each pixel, the average color indicator values are determined, which are assigned as a set of color indicators to the agricultural field for each multispectral image of the agricultural field in the second set;

- convert the values of a set of color indicators obtained at the previous stage using an encoder model into a fixed vector.

[0036] In another particular example of the method, the stage of determining soil moisture content indices for a given period of time for an agricultural field contains the steps of: determining the coordinates at which the soil moisture content was measured for a given period of time;

- determine the coordinates of the agricultural field

- assign soil moisture content indices to the agricultural field, obtained by triangulation based on the coordinates of the agricultural field and the coordinates where the soil moisture content was measured;

- convert the values of soil moisture content indices for a given period of time, obtained at the previous stage, using an encoder model into a fixed vector.

[0037] In another particular example of the method, the stage of determining soil moisture content indices for a given period of time for an agricultural field contains the stages of: determining the coordinates at which the soil moisture content was measured;

- determine the coordinates of the center of the agricultural field;

- determine the point closest to the center of the agricultural field at which the soil moisture content was measured;

- assign soil moisture content indices corresponding to the mentioned nearest point to the agricultural field;

[0038] In another particular example of the method, the step of determining weather data for a given period of time for an agricultural field contains steps in which: - obtain the coordinates of weather stations from which weather data were received for a given period of time;

- determine the coordinates of the agricultural field;

- assign weather data to the agricultural field, obtained by triangulation based on the coordinates of the agricultural field and the coordinates of weather stations;

- convert the values obtained at the previous stage using an encoder model into a fixed vector.

[0039] In another particular example of the method, the stage of determining weather data for a given period of time for an agricultural field contains stages in which:

- obtain the coordinates of weather stations from which weather data were received for a given period of time;

- determine the coordinates of the agricultural field;

- compare the coordinates of the agricultural field with the coordinates of weather stations and assign weather data for a given period of time to the nearest weather station to the agricultural field;

- convert the weather data values obtained at the previous stage using an encoder model into a fixed vector.

[0040] In another preferred embodiment of the claimed solution, a device for determining the yield of an agricultural field is presented, containing at least one computing device and at least one memory device containing machine-readable instructions, which, when executed by at least one computing device, perform the above way.

BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The features and advantages of the present technical solution will become apparent from the following detailed description of the invention and the accompanying drawings, in which:

[0042] In FIG. 1 - an example of data characterizing the dynamics of changes in the normalized vegetation index (NDVI) for a given agricultural field is presented, and the process of converting information into a vector of a fixed size is schematically depicted.

[0043] In FIG. 2 - an example of the architecture of a neural network of an encoder model is presented. [0044] In FIG. 3 - shows a diagram of the process of additional training of the yield determination model.

[0045] In FIG. 4 - an example of a data flow diagram is presented.

[0046] In FIG. 5 shows an example of a general view of a data processing system. [0047] In FIG. 6 shows an example of a general view of a computing device.

IMPLEMENTATION OF THE INVENTION

[0048] The concepts and terms necessary to understand this technical solution will be described below.

[0049] In this technical solution, a system means, including a computer system, a computer (electronic computer), CNC (computer numerical control), PLC (programmable logic controller), computerized control systems and any other devices capable of performing a given task. , a clearly defined sequence of operations (actions, instructions).

[0050] By command processing device is meant an electronic unit, a computing device, or an integrated circuit (microprocessor) that executes machine instructions (programs).

[0051] A command processing device reads and executes machine instructions (programs) from one or more storage devices. Storage devices can include, but are not limited to, hard drives (HDD), flash memory, ROM (read-only memory), solid-state drives (SSD), and optical drives.

[0052] A computing device is a counting and solving device that automatically performs one mathematical operation or a sequence of them in order to solve one problem or a class of similar problems (Great Soviet Encyclopedia. - M.: Soviet Encyclopedia. 1969 - 1978.).

[0053] Program - a sequence of instructions intended for execution by a computer control device or command processing device.

[0054] Software module - a program or a separate functional part of it, considered as a whole in the contexts of storage, replacement, translation, association with other program modules and its loading into the computer's RAM. [0055] Database (DB) - a collection of data organized according to a conceptual structure that describes the characteristics of that data and the relationships between them, a collection of data that supports one or more application areas (ISO/IEC 2382:2015, 2121423 "database").

[0056] A signal is a material embodiment of a message for use in the transmission, processing and storage of information.

[0057] Intensity is a scalar physical quantity that quantifies the power carried by a wave in the direction of propagation.

[0058] According to the diagram shown in FIG. 5, The data processing system contains:

- system 1 for obtaining multispectral images to determine light polarization values based on the image data;

- system 2 for obtaining multispectral images for determining, based on image data, pixel intensity values in various spectra;

- system 3 for determining soil moisture content;

- system 4 for determining weather data;

- user device 5;

- data preparation module 10;

- data management module 20;

- module 30 for determining agricultural field vegetation indices;

- module 40 for determining polarization indices;

- module 50 for determining spectral characteristics;

- module 60 for determining soil moisture;

- module 70 for determining weather data;

- module 80 for constructing numerical representations;

- module 90 for determining the yield of an agricultural field.

[0059] The multispectral image acquisition system 1 may be implemented on the basis of at least one server configured to receive multispectral imagery from at least one satellite. For example, satellites from the Sentinel-1 satellite constellation (https://sentinel.esa.int/web/sentinel/missions/sentinel-1) can be used as at least one satellite. Sentinel-1 is a series of Earth observation satellites developed by the European Space Agency agency. System 1 is not limited to a given constellation of satellites and can use data from similar constellations as an alternative.

[0060] The multispectral image acquisition system 2 may be implemented on the basis of at least one server configured to receive multispectral imagery from at least one satellite. For example, satellites from the Sentinel-2 satellite constellation (https://sentinel.esa.int/web/sentinel/missions/sentinel-2) can be used as at least one satellite. Sentinel-2 is an Earth observation mission from the Copernicus program that systematically acquires high spatial resolution (10 to 60 m) optical images over land and coastal waters. System 2 is not limited to a given constellation of satellites and can use data from similar constellations as an alternative.

[0061] The system 3 for obtaining multispectral images can be implemented on the basis of at least one server configured to receive multispectral images from at least one satellite designed for monitoring soil moisture content. For example, the at least one satellite may be a MetOp ASCAT satellite. MetOp (short for Meteorological Operational satellite) - a satellite weighing over four tons is designed to monitor the temperature and humidity of the Earth's surface, as well as monitor ozone levels and air flows over the ocean. System 3 is not limited to a given constellation of satellites and can use data from similar constellations as an alternative. Using the mentioned satellites, a third set of multispectral images can be obtained, on the basis of which indices of soil moisture content for a given period of time can be obtained using known methods (see, for example, the article by E.V. Vasilenko “Using data from the ascat/metop device for monitoring soil moisture ", Bulletin of Moscow University. Series 5. Geography, 2015).

[0062] The weather data determination system 4 can be implemented on the basis of local weather stations or satellites designed to collect weather data.

[0063] The data management module 20 may be implemented on at least one computing device capable of communicating data with the user device 5 and control the operation of the data processing system modules.

[0064] Module 30 for determining vegetation indices of an agricultural field can be implemented on the basis of a computing device capable of obtaining multispectral images to determine vegetation indices for each individual agricultural field.

[0065] Module 40 for determining polarization indices can be implemented on the basis of a computing device capable of obtaining multispectral images to determine polarization indices for each individual agricultural field.

[0066] The spectral characteristics determination module 50 may be implemented on the basis of a computing device capable of obtaining multispectral images to determine the spectral characteristics for each individual agricultural field.

[0067] The soil moisture determination module 60 may be implemented on the basis of a computing device capable of obtaining moisture data and determining soil moisture content indices for each individual agricultural field.

[0068] The weather data processing module 70 may be implemented on a computing device configured to determine weather data for each individual field.

[0069] Module 80 for constructing numerical representations can be implemented on a computing device and contain an encoder model pre-trained on a training data set to generate fixed vectors of a given length. Using this module, numerical representations can be obtained from the observed data series. A numerical representation is a mathematical mapping of a series of data, a set of characteristics, or any discrete structure into a vector of a given fixed size. It is believed that such a compressed representation contains all the necessary information that describes and distinguishes the object on which it was built. An example of such a mapping would be to represent the observed dynamics of the Normalized Vegetation Index (NDVI) into a vector consisting of 8 values. An example is shown in FIG. 1.

[0070] Encoder neural network. In order to convert the dynamics of the indices into a vector of a fixed length, that is, obtain a numerical representation, a trained encoder model is used. An encoder model is a neural network trained in such a way that it is capable of recreating the primary data sequence from a fixed-length vector. The architecture of the neural network is shown in Fig. 2.

As a specific implementation of an encoder neural network, it can consist of two LSTM (Long short-term memory) layers that operate on data sequences, and a layer with a fixed state between them. The parameters of the neural network are selected in such a way that it is able to restore the input sequence from a fixed state. In this case, the fixed state of the input sequence is its numerical representation.

[0071] Constructing numerical representations from vegetation indices.

In agriculture, various vegetation indices are widely used, reflecting the state of agricultural plantings.

Numerical representations from vegetation indices can be obtained based on information contained, for example, in images of the Sentinel-2 satellite group. The spectral characteristic may include: VIR - is the visible infrared spectrum, NIR - is the near-infrared spectrum, SWIR - short-wave infrared range.

The presented solution uses the following indices:

1. Cloudiness - the proportion of cloudy pixels inside the field. Information about the cloud coverage of images is stored in the image metadata.

2. Normalized vegetation index NDVI - an index for quantitative assessment of green vegetation. The range of NDVI values is -1 to 1. Negative NDVI values correspond to water. Values close to zero usually correspond to barren areas of rock, sand or snow. Low positive values correspond to shrubland and grassland (approximately 0.2 to 0.4), and high values correspond to temperate tropical rainforests.3. Difference vegetation index DVI is an index that distinguishes between soil and vegetation. Does not account for differences between reflectance and brightness caused by atmospheric effects and shadows. 4. Enhanced Vegetation Index Evi - an index that uses blue reflectance to correct background soil signals and reduce atmospheric disturbances. EVI values range from -1 to 1, with healthy vegetation typically 0.20 to 0.80.

5. Gemi Global Environmental Monitoring Index - This non-linear vegetation index is used for global environmental monitoring from satellite imagery. Aimed at correcting atmospheric effects. Similar to the NDVI index, but less sensitive to atmospheric influences.

6. Gari atmospheric green vegetation index - this index is more sensitive to a wide range of chlorophyll concentrations and less sensitive to atmospheric influences compared to NDVL

7. Green Difference Vegetation Index GDVI - The index was originally developed using color infrared photography to predict nitrogen requirements for corn.

8. GLI Leaf Greenness Index - GLI values range from -1 to 1. Negative values represent soil and non-living properties, while positive values represent green leaves and stems.

9. GOSAVI Optimized Soil Vegetation Index - This index was originally developed using color infrared photography to predict nitrogen requirements for corn. This index is similar to OSAVI, but replaces the green stripe with a red one.

10. GSAVI Soil Greenness Index - This index was originally developed using color infrared photography to predict nitrogen requirements for corn. This index is similar to SAVI, but replaces the green stripe with a red one.

11. Infrared vegetation index IPVI - this index is functionally the same as NDVI, but computationally faster. Values range from 0 to 1.

12. Leaf Area Index LAI - This index is used to assess leaf cover and also to predict crop growth and yield.

13. Modified non-linear vegetation index MNLI - this index is an improvement of the non-linear index (NLI). Includes Soil Adjusted Vegetation Index (SAVI) to account for soil background. ENVI uses a background correction factor (L) value of 0.5.

14. Modified Soil Adjusted Vegetation Index MSAVI2 - This index is a simpler version of the MSAVL Index Improves on the Soil Adjusted Vegetation Index (SAVI). This reduces soil noise and increases the dynamic range of the vegetation signal. MSAVI2 is based on an inductive method that does not use a constant L value to distinguish healthy vegetation.

15. Nonlinear vegetation index NLI - this index assumes that the relationship between many vegetation indices and surface biophysical parameters is nonlinear. It linearizes relationships with surface parameters that tend to be nonlinear.

16. Soil-Adjusted Vegetation Index OSAVI - This index is based on the Soil-Adjusted Vegetation Index (SAVI). It uses a standard value of 0.16 for the canopy background correction factor. This index is best used in areas with relatively sparse vegetation where the soil is visible through the canopy.

17. Renormalized Difference Vegetation Index RDVI - This index uses the difference between near-infrared and red wavelengths, as well as NDVI, to highlight healthy vegetation.

18. Soil Vegetation Index SAVI - The index is a conversion method that minimizes the influence of soil brightness from spectral vegetation indices using red and near-infrared (NIR) wavelengths.

19. Transformed difference vegetation index TDVI - this index is useful for monitoring vegetation cover in an urban environment.

20. Triangular vegetation index TGI - in this index, L represents the central wavelengths of the corresponding bands, p represents the pixel values of these bands. The TGI index is highly correlated with the chlorophyll content of leaves. TGI values are positive when the green reflectance is greater than the line between the red and blue vertices. This corresponds to green vegetation. TGI is negative when the green reflectance is less than the line between the red and blue vertices. This corresponds to features such as red soils. 21. Apparent Weathering Vegetation Index VARI - This index is based on ARVI and is used to estimate the proportion of vegetation in a scene with low weather sensitivity.

22. Normalized Difference Vegetation Index WDRVI - This index is similar to NDVI, but it uses a weighting factor a to reduce the discrepancy between the contributions of near-infrared and red signals to NDVI. WDRVI is especially effective in scenes with moderate to high vegetation density when NDVI exceeds 0.6.

For each field within the Technology, each of the above indices is calculated using Module 30. The indices are calculated for each pixel separately, then the index values are averaged within each field. Thus, for each field - for which the prediction is carried out - specified indices are determined for each of the available multispectral images. The set of indices is considered starting from September 1, until the moment when the classification is made. For example, to determine what yield is expected in a field as of July 21, 2021, indices are calculated from September 1, 2020 to July 21, 2021.

The input to Module 30 is images, for example, from a group of Sentinel-2 satellites. Each image represents one of 12 spectral channels for the selected area. Next, using the necessary spectral channels, it becomes possible to calculate vegetation indices for each pixel.

[0072] Accordingly, sequences of 22 indices for each of the available days are encoded in the Technology frame into a fixed vector of length 64. In this case, an encoder model is used, which is trained on the data of vegetation indices for the entire available observation period. Thus, according to the encoding results data of vegetation indices, for each agricultural field a vector of dimension 64 is obtained, reflecting the dynamics of the growing season.

[0073] Construction of numerical representations based on SAR channels.

As a specific implementation for obtaining SAR (Synthetic Aperture Radar) channels, the Sentinel-1 satellite constellation was used. Sentinel-1 can collect multiple different images from the same series of pulses by using its antenna to receive specific polarizations simultaneously. Sentinel-1 is a system built on the principle of radar synthesizing an aperture with dual polarization and phase conservation. It can transmit a signal with either horizontal (H) or vertical (V) polarization, and then receive a signal with both H and V polarization.

Objects on earth have distinctive polarization signatures, reflecting different polarizations at different intensities and converting one polarization to another. For example, volumetric scatterers (eg, a forest canopy) have different polarization properties than surface scatterers (eg, the sea surface).

The following indices are used within the Technology:

1. VH polarization;

2. Standard deviation of VH polarization for pixels within the field;

3. W polarization;

4. Standard deviation W of polarization for pixels inside the field;

5. Angle of incidence of the synthetic aperture radar beam

6. Standard deviation of the radar aperture synthesis beam incidence angle for pixels within the field. For each field within the Technology, each of the above indicators is considered. The indicators are calculated for each pixel separately, then the index values are averaged within each field. Thus, for each field - for which classification is carried out - specified parameters are determined for each of the available multispectral images. The set of parameters is considered starting from September 1, until the moment when the classification is made. For example, to determine what yield is expected in a field at the time of July 21, 2021, the parameters are calculated from September 1, 2020 to July 21, 2021.

Images of a group of Sentinel-1 satellites are supplied to the input of Module 40, each of which contains relevant information on VH polarization, W polarization and the angle of incidence of the synthetic aperture radar beam. Next, using the formula presented above, Module 40 calculates the corresponding standard deviations. These indicators are averaged for each field. The result is a correspondence between the numerical indicators and the corresponding fields.

[0074] Sequences of 6 parameters for each of the available days are encoded within the Technology into a fixed vector of length 16. In this case, an encoder model is used that is trained on these parameters for the entire available observation period. Thus, according to the coding results parameters, for each agricultural field a vector of dimension 16 is obtained, reflecting the dynamics of the growing season.

[0075] Constructing numerical representations based on optical channels.

As a specific implementation, the Sentinel-2 satellite group was used to obtain optical channels. Sentinel-2 is equipped with a multispectral imager (MSI). This sensor provides 13 spectral bands with pixel sizes ranging from 10 to 60 meters. There is no guarantee that the vegetation indices listed above store all the important information from the Sentinel-2 satellite data, therefore, within the Technology, primary channels are also used. The following channels are used within the Technology:

1. Ultra blue range (VIR), central wavelength - 443 nanometers.

2. Blue range (VIR), central wavelength - 490 nanometers.

3. Green range (VIR), central wavelength - 560 nanometers.

4. Red range (VIR), central wavelength - 665 nanometers.

5. Visible and near infrared light (NIR), central wavelength - 705 nanometers.

6. Visible and near infrared light (NIR), central wavelength - 740 nanometers.

7. Visible and near infrared light (NIR), central wavelength - 783 nanometers.

8. Visible and near infrared light (NIR), central wavelength - 842 nanometers.

9. Visible and near infrared light (NIR), central wavelength - 865 nanometers.

10. Short-wave infrared light (SWIR), central wavelength - 940 nanometers.

11. Short-wave infrared light (SWIR), central wavelength - 1375 nanometers.

12. Short-wave infrared light (SWIR), central wavelength - 1610 nanometers.

13. Short-wave infrared light (SWIR), central wavelength - 2190 nanometers.

14. Bit mask strip with cloud mask information. For each field within the Technology, each of the above indicators is considered. The indicators are calculated for each pixel separately, then the index values are averaged within each field. Thus, for each field - for which classification is carried out - specified parameters are determined for each of the available multispectral images. The set of parameters is considered starting from September 1, until the moment when the classification is made. For example, to determine what yield is expected in a field at the time of July 21, 2021, the parameters are calculated from September 1, 2020 to July 21, 2021.

Images in different resolutions and in different channels are supplied to the input of Module 50. In total, 14 different channels are downloaded to the same area of the Earth, that is, for the same area there are 14 different images with the corresponding channels.

Sequences of 14 parameters for each of the available days are encoded within the Technology into a fixed vector of length 64. In this case, an encoder model is used that is trained on these parameters for the entire available observation period. Thus, based on the results of encoding the parameters, for each agricultural field a vector of dimension 64 is obtained, reflecting the dynamics of the growing season.

[0076] Constructing numerical representations from soil moisture characteristics.

As part of the Technology, data on soil moisture content is used to improve the quality of forecasting. Soil moisture content data was determined using, for example, the MetOp ASCAT satellite.

The following data on soil moisture are used within the Technology:

1. moisture content index in the soil layer 7 - 28 cm. In a specific case, the indices can be calculated by the Italian Meteorological Institute (https://hsaf.meteoam.it/Products/ProductsList?type=soil_moisture product H14) according to the MetOp ASCAT satellite data, however Data received from other satellites can also be used. In general, the triangulation method can be used to correlate fields and their humidity. In a specific case, for each field the soil moisture content index indicator closest to the center of the agricultural field is determined. Subsequently, the soil moisture content index from this point is matched to the field under consideration. The sequence consisting of the humidity index for each of the available days is encoded within the Technology into a fixed vector of length 32. In this case, an encoder model is used that is trained on the humidity index for the entire available observation period. Thus, according to the coding results, for each agricultural field a vector of dimension 32 is obtained, reflecting the dynamics of soil moisture in the 7 - 28 cm layer.

[0077] Constructing numerical representations from weather data.

The Technology uses data from local weather stations, but data received from satellites can also be used. In general, the triangulation method can be used to correlate fields and weather data from weather stations. In a specific case, a series of weather data for a given agricultural field is determined by selecting the nearest point with available weather data using Module 70. After constructing a correspondence between the fields in question and available weather stations using the functionality of Module 70, the process of obtaining weather data for each field from its corresponding weather station occurs . Example, data is available from three weather stations: A, B and C. For a given agricultural field X, the geographically closest weather station is weather station B. Thus, the weather data for the agricultural field is used from weather station B.

The following weather data are used within the Technology:

1. atmospheric pressure;

2. relative air humidity;

3. wind speed;

4. gusts of wind;

5. dew point;

6. minimum soil temperature;

7. minimum air temperature;

8. average air temperature;

9. maximum air temperature;

10. standard deviation of air temperature;

11. precipitation;

12. snow depth.

The standard deviation of the X value is calculated using the following formula:

To calculate the standard deviation of air temperature, Xj is one air temperature measurement for a selected period of time, while d is the average of all air temperature measurements for a given period of time. N is the number of such measurements over a given period of time.

Sequences of 11 weather parameters for each of the available days are encoded within the Technology into a fixed vector of length 32. In this case, an encoder model is used, which is trained on weather parameters for the entire available observation period. The input to Module 80 is weather data, which is subsequently converted by Module 80 into a vector characterizing the weather data for the field under consideration. Thus, based on the results of encoding weather parameters, for each agricultural field a vector of dimension 32 is obtained, reflecting the weather dynamics.

[0078] Module 90 for determining the yield of an agricultural field can be implemented on the basis of a computing device and contain:

- module for training a model for predicting crop yields. This module includes a set of software for training a crop yield prediction model.

- module for obtaining predictions. This module includes a set of software for obtaining predictions for given fields and crops.

Based on the results of the work, the expected yield for a certain harvest year is determined for each field and the crop specified on it.

[0079] Module for training a crop yield prediction model.

A distinctive feature of the Technology is that it implements automatic additional training of the yield determination model. With one-time training, the problem of its portability and degradation of quality over time arises. An example is the impossibility of accurate forecasts in regions for which there was no training data, which differ significantly in conditions crop growing season, weather and soil cultivation technology.

The process of additional training in Module 80 occurs by loading additional data into the directory in which all training data is initially stored. Then, when you start retraining the Module 80 model, the new loaded data will be automatically added to the new final Module 80 training set.

[0080] The solution put forward within the Technology is to constantly retrain models when new data appears. Using the software, it becomes possible to effectively retrain models and thus maintain the quality of predictions. In FIG. Figure 3 schematically shows the mechanism for additional training of the model for determining the yield of agricultural fields. As new data becomes available, it is added to the overall training data corpus and the model is retrained. As a result of such additional training, the model’s ability to generalize increases.

[0081] Yield determination model.

Module 90 is equipped with a model designed to process data using the gradient-growing tree method in order to determine the value of the expected yield of an agricultural field. The method was chosen because it provides the best quality based on cross-validation results. A separate model is trained for each crop. The values of learning rate, tree configuration, set of predictors, and other method parameters were selected as hyperparameters. Hyperparameters were selected at the same time for each agricultural crop separately and do not change as part of additional training.

[0082] Model training.

Models are trained using the following data:

1. numerical representations of vegetation indices;

2. numerical representations of SAR channels;

3. numerical representations of optical channels;

4. numerical representations of soil moisture characteristics;

5. numerical representations of weather data; 6. history of crop rotation, that is, what was planted before and what the yield was;

7. information about what agricultural crop is planted;

8. real data on the yield of agricultural fields.

From all the data, a training corpus is formed, which is used to build a model for each crop. The training process is an iterative optimization with the goal of obtaining a minimum error on the training data. Training is carried out in such a way that the model is able to determine the yield at any time during the sowing season. This is achieved by including in the training corpus examples that are trimmed in terms of the composition of the coded values. For example, the corpus includes data in which all sequences are calculated not at the end of the season, but at May 1, June 1, and so on.

[0083] Prediction module.

The prediction module is a set of software that, based on prepared data, determines the yield of an agricultural crop. The module uses the following data:

1. numerical representations of vegetation indices;

2. numerical representations of SAR channels;

3. numerical representations of optical channels;

4. numerical representations of soil moisture characteristics;

5. numerical representations of weather data;

6. history of crop rotation, that is, what was planted before and what the yield was;

7. information about the type of crop for which the expected yield value is determined.

Preprocessing and data acquisition occurs in accordance with modules 30, 40, 50, 60 and 70, respectively. All data is obtained for a specific set of fields that are supplied as input. For each such field its coordinates are known. Next, the preprocessed data is combined into a single data frame based on the time the observations were received. After preprocessing, Module 80 generates a ready-made dataframe with the names of the fields and the corresponding data. This dataframe is input to the Module 80 model for yield forecasting. The output generates yield predictions for each agricultural field.

[0084] Data flow diagram.

A data flow diagram describing the format and type of information at each step is shown in Figure 4. The following formats and data types are used step by step:

1. The operator transmits all information on agricultural fields in csv file format to the input of the module for constructing numerical representations.

2. Based on the encoding results, data is generated with numerical representations of indices, weather, channels, etc.

3. The prediction module, using a set of trained models for each crop, determines the yield for each individual field. The results of the work are generated for the operator in a csv file. [0085] The data processing system operates as follows.

[0086] In the first stage of operation of the data processing system, the data management module 20 receives a request to determine the yield of a field. Said request may be sent, for example, through a user device 5, which may be a laptop or desktop computer, telephone, smartphone, tablet or other computing device equipped with wired and/or wireless communication means designed to exchange data with said module 20.

[0087] To generate a request to determine the yield of a field, the data management module 20, after accessing it from the device 5, can provide access to information about a set of agricultural fields for which the yield can be determined. For example, said information about a set of agricultural fields can be presented to the user of the device 5 in a graphical user interface, in which the user can also select at least one agricultural field of interest to the user, and the type of agricultural crop for which it is necessary to determine the yield .

[0088] In an alternative embodiment of the presented solution, the user of device 5 can generate a text file and indicate in it the type of agricultural crop and the coordinates of at least one agricultural field for which the yield should be determined, after which device 5 will generate a request to determine the yield of an agricultural field, containing the mentioned text file. Additionally, in the request to determine the yield of an agricultural field, the user of device 5 can include data characterizing the history of crop rotation and the actual yield for a given period of time. [0089] Accordingly, the field yield determination request received by the data management module 20 may include an agricultural field ID and a crop type ID. Said module 20 retrieves the agricultural field ID from the received request and accesses the DB 10 to retrieve data related to said agricultural field ID. In particular, said module 20 retrieves:

- coordinates of the agricultural field;

- coordinates of the center of the agricultural field;

- the first set of multispectral images for a period of time specified by the developer of the module 20, and each multispectral image of an agricultural field contains light polarization values, in particular the values of VH polarization, W polarization and the angle of incidence of the synthetic aperture radar beam;

- a second set of multispectral images for a period of time specified by the developer of the module 20, and each multispectral image of an agricultural field contains pixel intensity values in the visible infrared spectrum, near-infrared spectrum and short-wave infrared spectrum, as well as cloudiness values of image pixels that can be represented, for example, in the form of a binary mask, the values of which indicate whether a pixel in an agricultural field is cloudy or not;

- indices of soil moisture content for a period of time specified by the developer of module 20, tied to the coordinates in which the soil moisture content was measured, for example, in a layer of 7 - 28 cm;

- weather data from weather stations for a period of time specified by the developer of module 20, linked to the coordinates of weather stations, including indicators: atmospheric pressure; relative air humidity; wind speed; gusts of wind; dew point; minimum soil temperature; minimum air temperature; average air temperature; maximum air temperature; standard deviation of air temperature; precipitation; snow depth.

[0090] Next, the second set of multispectral images of the agricultural field is sent by the mentioned module 20 to the module 30 for determining the vegetation indices of the agricultural field, which, based on the received multispectral image data, determines for each image a set of agricultural field vegetation indices. The number of vegetation indices included in the mentioned set can be preset by the developer of module 30. The set of agricultural field vegetation indices for each multispectral image of an agricultural field from the second set is determined as follows.

[0091] To determine the cloudiness index of an agricultural field, the mentioned module 30 extracts from the received multispectral image data the cloudiness values of the image pixels, and then, based on the number of cloudy pixels, determines the mentioned cloudiness index of the agricultural field, for example, as the ratio of the number of cloudy pixels to the total to the number of pixels in the agricultural field.

[0092] To determine the normalized vegetation index (NDVI), the mentioned module 30 for each pixel extracts from the received multispectral image data the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range and based on them determines the NDVL value for each pixel For example, the mentioned value can be determined by the formula:

Next, the mentioned module 30, based on the NDVI values of the pixels, determines the average NDVI value, which is assigned as the NDVI index to the agricultural field. [0093] Also based on the NIR pixel intensity value and the red pixel intensity value, said module 30 determines difference vegetation index (DVI) values for each pixel of the agricultural field. For example, the DVI value can be determined by the formula:

DVI = NIR - Red

Next, said module 30 determines, based on the DVI values of the pixels, an average DVI value, which is assigned as a DVI index to the agricultural field.

[0094] To determine the enhanced vegetation index (EVI), the mentioned module 30, in addition to the NIR pixel intensity value and the red pixel intensity value, also extracts the blue pixel intensity value for each pixel, after which, based on them, the mentioned module 30 for each pixel is determined by the EVL value. For example, the EVI value can be determined by the formula: EVI-2.5

Next, said module 30 determines, based on the EVI values of the pixels, the average EVI value, which is assigned as the EVI index to the agricultural field.

[0095] The global environmental monitoring index (Gemi) value for each pixel of the agricultural field image by said module 30 can also be determined based on the pixel intensity value in the near infrared spectrum (NIR) and the pixel intensity value in the red range, for example, by formula:

Red - 0.125

GEMI = eta( vl - 0.25

Where:

2(NIR ² - Red ² ) + 1.5 * NIR + 0.5 * Red eta = -

NIR + Red + 0.5

Next, the mentioned module 30, based on the Gemi values of the pixels, determines the average Gemi value, which is assigned as the Gemi index to the agricultural field. [0096] To determine the green vegetation index resistant to atmospheric influences (Gari), the mentioned module 30 for each pixel extracts from the acquired multispectral image data a pixel intensity value in the near infrared spectrum (NIR), a pixel intensity value in the blue range, a pixel intensity value in the green range and the pixel intensity value in the red range and based on them, the Gari value is determined for each pixel. For example, the mentioned value can be determined by the formula:

Next, the mentioned module 30, based on the Gari values of the pixels, determines the average Gari value, which is assigned as the Gari index to the agricultural field.

[0097] To determine the green difference vegetation index (GDVI), the mentioned module 30 for each pixel extracts from the acquired multispectral image data the pixel intensity value in the near infrared spectrum (NIR) and the pixel intensity value in the green range and based on them determines for each pixel GDVI value. For example, the mentioned value can be determined by the formula:

GDVI = NIR - Green Next, the mentioned module 30, based on the GDVI values of the pixels, determines the average GDVI value, which is assigned as the Gari index to the agricultural field.

[0098] To determine the leaf greenness index (GLI), the module 30, for each pixel, extracts from the acquired multispectral image data a pixel intensity value in the blue range, a pixel intensity value in the green range and a pixel intensity value in the red range and based on them for each pixel defines the GLL value For example, the mentioned value can be determined by the formula:

Next, the mentioned module 30, based on the GLI values of the pixels, determines the average GLI value, which is assigned as the GLI index to the agricultural field.

[0099] To determine the optimized soil vegetation index (GOSAVI), the mentioned module 30 for each pixel extracts from the acquired multispectral image data the pixel intensity value in the near infrared spectrum (NIR) and the pixel intensity value in the green range and based on them determines the value for each pixel GOSAVI. For example, the mentioned value can be determined by the formula:

NIR - Green

GOSAVI = -

NIR + Green + 0.16

Next, the mentioned module 30, based on the GOSAVI values of the pixels, determines the average GOSAVI value, which is assigned as the GOSAVI index to the agricultural field.

[0100] To determine the soil greenness index (GSAVI), the mentioned module 30 for each pixel extracts from the obtained multispectral image data the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the green range and based on them determines the GSAVI value for each pixel . For example, the mentioned value can be determined by the formula:

GSAVI = 1.5

Next, the mentioned module 30, based on the GSAVI values of the pixels, determines the average GSAVI value, which is assigned as the GSAVI index to the agricultural field.

[0101] To determine the infrared vegetation index (IPVI), said module 30 for each pixel extracts from the received data multispectral image, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range and based on them for each pixel determines the IPVL value. For example, the mentioned value can be determined by the formula:

NIR

IPVI = -

NIR+Red

Next, the mentioned module 30, based on the IPVI values of the pixels, determines the average IPVI value, which is assigned as the IPVI index to the agricultural field.

[0102] The leaf area index (LAI) of the agricultural field by said module 30 can be determined based on the EVI value of the agricultural field, for example, by the formula:

LAI = (3.618 * EVI - 0.118)

[0103] To determine the modified non-linear vegetation index (MNLI), the mentioned module 30 for each pixel extracts from the obtained multispectral image data the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range and based on them determines the value for each pixel MNLI.

For example, the mentioned value can be determined by the formula:

where L is the value of the background correction factor specified by the developer of module 30, which can be, for example, 0.5.

Next, the mentioned module 30, based on the MNLI values of the pixels, determines the average MNLI value, which is assigned as the MNLI index to the agricultural field. [0104] To determine the Modified Soil Corrected Vegetation Index (MSAVI2), said module 30, for each pixel, extracts from the acquired multispectral image data a near-infrared (NIR) pixel intensity value and a red pixel intensity value and based thereon for each pixel is determined by the MSAVI2 value. For example, the mentioned value can be determined by the formula:

Next, the mentioned module 30, based on the MSAVI2 pixel values, determines the average MSAVI2 value, which is assigned as the MSAVI2 index to the agricultural field. [0105] To determine the nonlinear vegetation index (NLI), the mentioned module 30 for each pixel extracts from the received multispectral image data the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range and based on them determines the NLL value for each pixel For example, the mentioned value can be determined by the formula:

NIR ² - Red N11 - _

NIR ² + Red

Next, the mentioned module 30, based on the NLI values of the pixels, determines the average NLI value, which is assigned as the NLI index to the agricultural field.

[0106] To determine the Optimized Soil Sensitive Vegetation Index (OSAVI), said module 30 extracts from the acquired multispectral image data a near-infrared (NIR) pixel intensity value and a red pixel intensity value for each pixel and based thereon on each pixel. defines the OSAVI value. For example, the mentioned value can be determined by the formula:

Next, the mentioned module 30, based on the OSAVI pixel values, determines the average OSAVI value, which is assigned as the OSAVI agricultural index

FIELD.

[0107] To determine the renormalized difference vegetation index (RDVI), the mentioned module 30 for each pixel extracts from the received multispectral image data the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range and based on them determines the value for each pixel RDVL

For example, the mentioned value can be determined by the formula:

Next, the mentioned module 30, based on the RDVI values of the pixels, determines the average RDVI value, which is assigned as the RDVI index to the agricultural field.

[0108] To determine the soil vegetation index (SAVI), the module 30, for each pixel, extracts from the acquired multispectral image data a pixel intensity value in the near infrared spectrum (NIR) and a pixel intensity value in the red spectrum range and, based on them, determines the SAVI value for each pixel. For example, the mentioned value can be determined by the formula:

1.5 SAVI =

(NI

Next, the mentioned module 30, based on the SAVI values of the pixels, determines the average SAVI value, which is assigned as the SAVI index to the agricultural field. [0109] To determine the transformed difference vegetation index (TDVI), the mentioned module 30 for each pixel extracts from the received multispectral image data the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range and based on them determines the value for each pixel TDVL For example, the mentioned value can be determined by the formula: ]

0.5

Next, the mentioned module 30, based on the TDVI values of the pixels, determines the average TDVI value, which is assigned as the TDVI index to the agricultural field. [0110] To determine the triangular vegetation index (TGI), the module 30 extracts from the acquired multispectral image data a blue pixel intensity value, a green pixel intensity value, and a red pixel intensity value for each pixel and based thereon on each pixel. defines the TGL value For example, the mentioned value can be determined by the formula:

Said parameters for determining the TGI value can be obtained by methods known in the art, for example those disclosed in Hunt, E., S. Daughtry, J. Eitel, and D. Long, “Remote Sensing Leaf Chlorophyll Content Using a Visible Band Index." Agronomy Journal 103, no. 4 (2011): 1090-1099. Next, the mentioned module 30, based on the TGI values of the pixels, determines the average TGI value, which is assigned as the TGI index to the agricultural field. [0111] To determine the Visible Weather Resistant Vegetation Index (VARI), said module 30, for each pixel, extracts from the acquired multispectral image data a blue pixel intensity value, a green pixel intensity value, and a pixel intensity in the red range and, based on them, determines the VARI value for each pixel. For example, the mentioned value can be determined by the formula:

Green - Red

VARI = - — — — —

Green + Red - Blue

5 Next, the mentioned module 30, based on the VARI values of the pixels, determines the average VARI value, which is assigned as the VARI index to the agricultural field.

[0112] To determine the normalized difference vegetation index (WDRVI), the mentioned module 30 for each pixel extracts the pixel intensity value in the near range from the acquired multispectral image data

10 infrared spectrum (NIR) and the pixel intensity value in the red range and based on them, the WDRVI value is determined for each pixel. For example, the mentioned value can be determined by the formula:

Next, the mentioned module 30, based on the WDRVI values of the pixels, determines the average WDRVI value, which is assigned as the WDRVI index to the agricultural field.

[0113] Thus, module 30 for determining agricultural field vegetation indices for each received multispectral image determines and generates a set of agricultural field vegetation indices, which includes all 22

20 index defined by the mentioned module 30 above.

[0114] The resulting sets of vegetation indices are sent by the mentioned module 30 to the module 80 for constructing numerical representations.

[0115] Also, the second set of multispectral images of the agricultural field is sent by the mentioned module 20 to the module 50 for determining spectral characteristics, which, based on the received multispectral image data, determines a set of color parameters of the agricultural field for each image. The number of color parameters included in the said set can be predetermined by the developer of the module 50. The set of color parameters for each image from the second set is determined as follows. zo [0116] To determine a set of color indicators, the mentioned module 50 extracts from the received data for each pixel:

- pixel intensity value in the ultra blue range, obtained at a resolution of 60 meters with a central wavelength of 443 nanometers; - pixel intensity value in the blue range, obtained at a resolution of 10 meters with a central wavelength of 490 nanometers;

- pixel intensity value in short-wave infrared light (SWIR), obtained at a resolution of 20 meters with a central wavelength of 2190 nanometers; - a bitmask strip with information about the cloud mask, representing the cloudiness values of the pixels of the agricultural field image, where 1 is responsible for the presence of clouds and 0 for the absence.

[0117] Next, the mentioned module 50, based on the mentioned color indicator values extracted for each pixel, determines the average color indicator values, which are assigned as a set of color indicators to the agricultural field. Accordingly, a set of color indicators is determined for each image from the second set, after which the resulting sets of color indicators are sent by the mentioned module 50 to the module 80 for constructing numerical representations.

[0118] The first set of multispectral images of the agricultural field is sent by the data management module 20 to the polarization indices determination module 40, which determines for each multispectral image in the set a set of polarization indices of the agricultural field. The number of polarization indices included in the said set can be predetermined by the developer of the module 40. The set of polarization indices of the agricultural field for each image in the set can be determined as follows. In particular, the mentioned module 40 extracts from the received image data for each pixel of the agricultural field the values of VH polarization, W polarization and the angle of incidence of the radar aperture synthesis beam and, on their basis, determines for each pixel:

- value of standard deviation VH polarization;

- value of standard deviation W of polarization;

- the value of the standard deviation of the angle of incidence of the radar aperture synthesis beam.

[0119] For example, to determine the polarization standard deviation value VH for each picture, said module 40, based on the polarization VH values of each agricultural pixel, determines the average polarization value VH. Also, the mentioned module 40 determines for each image, based on the polarization values W of each agricultural pixel, the average value W of the polarization and, based on the values of the angle of incidence of the beam of each agricultural pixel, the average value of the angle of incidence of the synthetic aperture radar beam

[0120] Next, based on the average polarization value VH, the polarization value VH of each pixel of the agricultural field and the total number of pixels of the agricultural field, said module 40 determines the value of the standard deviation VH polarization. For example, the value of the standard deviation of VH polarization can be determined by the formula:

where d is the average value of VH polarization;

- value VH of polarization of i-ro pixel in the considered agricultural field; n is the value of the total number of pixels.

[0121] Similarly, the mentioned module 40 can be defined:

- the value of the standard deviation W of polarization based on the average value of VH polarization, the value of VH polarization of each pixel of the agricultural field and the total number of pixels of the agricultural field;

- the value of the standard deviation of the angle of incidence of the radar aperture synthesis beam based on the average value of the angle of incidence of the beam, the value of the angle of incidence of the beam of each pixel of the agricultural field and the total number of pixels of the agricultural field.

[0122] Next, the mentioned module 40 assigns the obtained average values of VH polarization, W polarization, angle of incidence of the beam and the above-determined deviation values as polarization indices to the agricultural field. Thus, module 40 for determining polarization indices for each multispectral image in the set determines and generates a set of polarization indices for the agricultural field, which includes the polarization indices defined above. The resulting sets of polarization indices are sent by the mentioned module 40 to the module 80 for constructing numerical representations.

[0123] Information about the soil moisture content indices for a given period of time, linked to the coordinates in which the soil moisture content was measured, as well as the coordinates of the center of the agricultural field, the data control module 20 transmits to the soil moisture determination module 60, which, by using the method nearest neighbors determines the point closest to the center of the agricultural field at which the soil moisture content was measured, and assigns the soil moisture content indices corresponding to the mentioned nearest point to the agricultural field.

[0124] In an alternative embodiment of the presented solution, said soil moisture content indices can be assigned to an agricultural field by said module 60 by processing data on the coordinates of the agricultural field and the coordinates where soil moisture content measurements were taken using the method triangulation. The triangulation method is widely known in the art and will not be described in more detail in this application. The resulting indices of soil moisture content for a given period of time are sent by the mentioned module 60 to the module 80 for constructing numerical representations.

[0125] Weather data from weather stations for a given period of time, linked to the coordinates of the weather stations, as well as the coordinates of the agricultural field, the data control module 20 transmits to the weather data determination module 70, which compares the coordinates of the agricultural field with the coordinates of the weather stations and assigns x field weather data from the nearest weather station.

[0126] In an alternative embodiment of the presented solution, said weather data can be assigned to an agricultural field by said module 70 by processing data on the coordinates of the agricultural field and the coordinates of weather stations using the triangulation method. The triangulation method is widely known in the art and will not be described in more detail in this application. The assigned weather data for a given period of time is sent by said module 70 to the numerical representation module 80.

[0127] The data obtained by the module 80 for constructing numerical representations is converted by known methods into vectors of a fixed length, for example:

- a set of vegetation indices can be transformed into a fixed vector of dimension 64, reflecting the dynamics of vegetation for a given period of time;

- a set of color indicators can be converted into a fixed vector of dimension 64, reflecting the dynamics of the growing season for a given period of time;

- a set of polarization indices can be transformed into a fixed vector of dimension 16, reflecting the dynamics of the growing season for a given period of time;

- the soil moisture content index can be converted into a fixed vector of dimension 32, reflecting the dynamics of soil moisture over a given period of time;

- weather data can be converted into a fixed vector of dimension 32, reflecting the weather dynamics for a given period of time.

[0128] The received vectors by module 80 are sent to module 90 for determining the yield of an agricultural field, which, based on them and on the basis of data characterizing the type of crop received from module 20 determines the yield value of the agricultural field, for example, the expected yield for the year for the agricultural field selected by the user. When determining yield, module 90 can also take into account data characterizing crop rotation history and actual yield over a given period of time. Said data may be provided by the user of Device 5 and retrieved by module 90 from a yield determination request, or uploaded directly by the developer into module 90, or provided by an alternative known method.

[0129] To obtain the yield value of an agricultural field, the mentioned module 90, based on data characterizing the type of crop, selects a model previously trained on the training data set and designed to predict the yield for a given type of crop, and then extracts it from the data characterizing the history of crop rotation and actual yield, yield values tied to a time value, and feeds them, along with fixed-length vectors received from module 80, to the model input. At the output of the model, module 90 receives the value of the yield of the agricultural field, in particular, the expected yield.

[0130] The resulting expected yield value is sent by module 90 to module 20 for display to the user of device 5.

[0131] In general (see Fig. 6), a computing device (200) contains one or more processors (201), memory devices such as RAM (202) and ROM (203), and input/ output (204), input/output devices (205), and network communication device (206).

[0132] The processor (201) (or multiple processors, multi-core processor, etc.) may be selected from a variety of devices commonly used today, for example, from manufacturers such as: Intel™, AMD™, Apple™, Samsung Exynos ™, MediaTEK™, Qualcomm Snapdragon™, etc. The processor or one of the processors used in the system (200) must also include a graphics processor, for example an NVIDIA GPU with a CUDA-compatible programming model or Graphcore, the type of which is also suitable for carrying out the method in whole or in part, and can also be used for training and application of machine learning models in various information systems. [0133] RAM (202) is a random access memory and is designed to store machine-readable instructions executed by the processor (201) to perform the necessary logical data processing operations. The RAM (202) typically contains executable operating system instructions and associated software components (applications, program modules, etc.). In this case, the available memory capacity of the graphics card or graphics processor can act as RAM (202).

[0134] The ROM (203) is one or more permanent storage devices, such as a hard disk drive (HDD), a solid state drive (SSD), flash memory (EEPROM, NAND, etc.), optical storage media ( CD-R/RW, DVD-R/RW, BlueRay Disc, MD), etc.

[0135] To organize the operation of device components (200) and organize the operation of external connected devices, various types of I/O interfaces (204) are used. The choice of appropriate interfaces depends on the specific design of the computing device, which can be, but is not limited to: PCI, AGP, PS/2, IrDa, FireWire, LPT, COM, SATA, IDE, Lightning, USB (2.0, 3.0, 3.1, micro, mini, type C), TRS/Audio jack (2.5, 3.5, 6.35), HDMI, DVI, VGA, Display Port, RJ45, RS232, etc.

[0136] To ensure user interaction with the computing device (200), various means (205) of I/O information are used, for example, a keyboard, a display (monitor), a touch display, a touch pad, a joystick, a mouse, a light pen, a stylus, touch panel, trackball, speakers, microphone, augmented reality tools, optical sensors, tablet, light indicators, projector, camera, biometric identification tools (retina scanner, fingerprint scanner, voice recognition module), etc.

[0137] The network communication means (206) provides data transmission via an internal or external computer network, for example, an Intranet, the Internet, a LAN, etc. One or more means (206) may be used, but not limited to: Ethernet card, GSM modem, GPRS modem, LTE modem, 5G modem, satellite communication module, NFC module, Bluetooth and/or BLE module, Wi-Fi module and etc.

[0138] Additionally, satellite navigation tools can also be used as part of the device (200), for example, GPS, GLONASS, BeiDou, Galileo. [0139] The specific selection of device elements (200) for implementing various software and hardware architectural solutions may vary while maintaining the required functionality provided.

[0140] Modifications and improvements to the above-described embodiments of the present technical solution will be apparent to those skilled in the art. The foregoing description is provided by way of example only and is not intended to be limiting. Thus, the scope of the present technical solution is limited only by the scope of the attached claims.

Claims

Claim.

1. A method for determining the yield of an agricultural (agricultural) field, performed by at least one computing device, containing the steps of:

- obtain the first set of multispectral images for a given period of time, and each multispectral image of an agricultural field contains values of light polarization, in particular the values of VH polarization, W polarization and the angle of incidence of the synthetic aperture radar beam;

- based on the mentioned first set of images, a set of polarization indices for the agricultural field is determined;

- determine weather data for a given period of time for an agricultural field;

2. The method according to claim 1, characterized in that the determination of the yield of an agricultural field is carried out taking into account data characterizing the history of crop rotation and the actual yield for a given period of time.

3. The method according to claim 1, characterized in that soil moisture content indices for a given period of time for an agricultural field are determined based on data from the third set of multispectral images for a given period of time, obtained using at least one satellite designed for monitoring moisture content soil.

44

4. The method according to claim 1, characterized in that the stage of determining a set of polarization indices for an agricultural field contains stages in which:

- determined for each pixel of the agricultural field based on the extracted values: the value of the standard deviation of VH polarization; polarization standard deviation value W; the value of the standard deviation of the angle of incidence of the radar aperture synthesis beam;

- determine the average values of VH polarization, W polarization, angle of incidence of the beam and the deviation values defined above;

- assign average values of VH polarization, W polarization, angle of incidence of the beam and the deviation values defined above as polarization indices to the agricultural field for each multispectral image of the agricultural field in the first set;

5. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the value of the normalized vegetation index (NDVI) is determined for each pixel;

6. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

45

7. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR), the pixel intensity value in the red range and the pixel intensity value in the blue range are extracted, after which, based on them, the pixel intensity value is determined for each pixel improved vegetation index (EVI) value;

- determine the leaf surface index (LAI) based on the EVI index value.

8. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

9. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR), the pixel intensity value in the blue range, the pixel intensity value in the green range and the pixel intensity value in the red range are extracted and on them on a basis for each pixel, the value of the green vegetation index resistant to atmospheric influences (Gari) is determined;

101. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near field is extracted

46 infrared spectrum (NIR) and the pixel intensity value in the green range and based on them, the green difference vegetation index (GDVI) value is determined for each pixel;

11. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the blue range, the pixel intensity value in the green range and the pixel intensity value in the red range are extracted and on their basis the value of the leaf greenness index (GLI) is determined for each pixel );

12. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

13. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

14. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which: - from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the infrared vegetation index (IPVI) value is determined for each pixel;

15. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the value of the modified nonlinear vegetation index (MNLI) is determined for each pixel;

16. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

17. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

18. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

19. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the value of the renormalized difference vegetation index (RDVI) is determined for each pixel;

20. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

21. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- from the obtained data of the second set of multispectral images for each pixel of the agricultural field, the pixel intensity value in the near-infrared spectrum (NIR) and the pixel intensity value in the red range are extracted and on their basis the value of the transformed difference vegetation index (TDVI) is determined for each pixel; - based on the TDVI values of the pixels, the average TDVI value is determined, which is assigned as the TDVI index to the agricultural field.

22. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

- based on the TGI values of the pixels, the average TGI value is determined, which is assigned as the TGI index to the agricultural field.

23. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

24. The method according to claim 1, characterized in that the stage of determining a set of vegetation indices for an agricultural field contains stages in which:

25. The method according to claim 1, characterized in that additional steps are performed in which:

50 - based on the number of cloudy pixels, the cloudiness index of the agricultural field is determined.

26. The method according to claim 1, characterized in that an additional step is performed in which:

- convert the obtained values of vegetation indices assigned to the agricultural field using an encoder model into a fixed vector.

27. The method according to claim 1, characterized in that the stage of determining a set of color indicators for an agricultural field contains stages in which: the following is extracted from the received data for each pixel:

- pixel intensity value in the ultra blue range, obtained at a resolution of 60 meters with a central wavelength of 443 nanometers;

- pixel intensity value in the blue range, obtained at a resolution of 10 meters with a central wavelength of 490 nanometers;

51 - pixel intensity value in short-wave infrared light (SWIR), obtained at a resolution of 60 meters with a central wavelength of 1375 nanometers;

- cloudiness values of agricultural field image pixels;

- based on the above-mentioned color indicator values extracted for each pixel, the average color indicator values are determined, which are assigned as a set of color indicators to the agricultural field for each multispectral image of the agricultural field in the second set;

28. The method according to claim 1, characterized in that the stage of determining soil moisture content indices for a given period of time for an agricultural field contains stages in which: determine the coordinates at which the soil moisture content was measured for a given period of time;

- determine the coordinates of the agricultural field;

29. The method according to claim 1, characterized in that the stage of determining soil moisture content indices for a given period of time for an agricultural field contains stages in which: determine the coordinates at which the soil moisture content was measured;

- determine the coordinates of the center of the agricultural field;

52 - determine the point closest to the center of the agricultural field at which the soil moisture content was measured;

- assign soil moisture content indices corresponding to the mentioned nearest point to the agricultural field.

- convert the values obtained at the previous stage using an encoder model into a fixed vector

30. The method according to claim 1, characterized in that the stage of determining weather data for a given period of time for an agricultural field contains stages in which:

- receive the coordinates of weather stations from which weather data was received;

- determine the coordinates of the agricultural field;

31. The method according to claim 1, characterized in that the stage of determining weather data for a given period of time for an agricultural field contains stages in which:

- determine the coordinates of the agricultural field;

- compare the coordinates of the agricultural field with the coordinates of weather stations and assign the weather data of the nearest weather station to the agricultural field.

- convert weather data values for a given period of time, obtained at the previous stage, using an encoder model into a fixed vector

32. A device for determining the yield of an agricultural field, containing at least one computing device and at least one memory device containing machine-readable instructions, which, when executed by at least one computing device, perform the method according to any one of claims. 1- 31.

53