A method, apparatus, medium, device and product for enhancing soil moisture data

Abstract

The application belongs to the technical field of earth science, and particularly relates to a soil moisture data enhancement method, device, medium, equipment and product. The application obtains satellite remote sensing soil moisture data, land surface assimilation soil moisture data and environment data of a target region within a certain time range; determines a soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within a certain time range; determines a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map; and determines enhanced soil moisture data with resolution enhancement according to the soil moisture multi-source fusion feature map and the environment data. The obtained enhanced soil moisture data has high consistency with site measured soil moisture, and can provide support of soil moisture data with time and space continuity and high resolution for related hydro-meteorological research.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of earth science, and particularly relates to a soil moisture data enhancement method, device, medium, equipment and product. BACKGROUND

[0002] The soil moisture retrieved from microwave radiation brightness temperature has high accuracy, but is discontinuous in time and space and has low spatial resolution, and is difficult to meet the demand of fine soil moisture monitoring. Deep learning is a large deep artificial neural network tool that has attracted much attention in recent years, and a multi-level network can deeply mine multi-scale spatial features and potential correlations of various environmental parameters, and is widely applied to solving various problems in the field of hydrology. However, there is still a lack of research on retrieving high-resolution and time-space continuous soil moisture by using deep learning. SUMMARY

[0003] To solve the above technical problems, the application provides a soil moisture data enhancement method, device, medium, equipment and product. The application obtains satellite remote sensing soil moisture data, land surface assimilation soil moisture data and environmental data of a target region within a certain time range; determines a soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within the certain time range; determines a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map; and determines resolution-enhanced enhanced soil moisture data according to the soil moisture multi-source fusion feature map and the environmental data. The obtained enhanced soil moisture data has high consistency with the site measured soil moisture, and can provide time-space continuous and high-resolution soil moisture data support for related hydro-meteorological research.

[0004] To solve the above technical problems, the application provides a technical solution including five aspects.

[0005] In a first aspect, the application provides a soil moisture data enhancement method, including: obtaining satellite remote sensing soil moisture data, land surface assimilation soil moisture data and environmental data of a target region within a certain time range; determining a soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within the certain time range; determining a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map; and determining resolution-enhanced enhanced soil moisture data according to the soil moisture multi-source fusion feature map and the environmental data.

[0006] In some embodiments, the determining the soil moisture reconstruction feature map of the target area according to the satellite remote sensing soil moisture data and the land surface assimilated soil moisture data within a certain time range comprises: extracting effective spatial features of each time phase from the satellite remote sensing soil moisture data through a partial convolution layer; obtaining time phase information of each effective spatial feature; fusing the effective spatial features of each time phase according to the time phase information to obtain a microwave soil moisture continuous feature map of the target area; determining an assimilated soil moisture feature map of the target area according to the land surface assimilated soil moisture data; and determining the soil moisture reconstruction feature map according to the assimilated soil moisture feature map and the microwave soil moisture continuous feature map.

[0007] In some embodiments, the determining the soil moisture multi-source fusion feature map according to the land surface assimilated soil moisture data and the soil moisture reconstruction feature map comprises: determining soil moisture multi-level features of the target area according to the land surface assimilated soil moisture data; determining soil moisture multi-scale features of the target area according to the soil moisture multi-level features; and determining the soil moisture multi-source fusion feature map according to the soil moisture multi-scale features and the soil moisture reconstruction feature map.

[0008] In some embodiments, the determining the resolution-enhanced enhanced soil moisture data according to the soil moisture multi-source fusion feature map and the environmental data comprises: determining environmental features of the target area according to the environmental data; performing resolution enhancement on the soil moisture multi-source fusion feature map according to the environmental features to obtain a soil moisture resolution-enhanced feature map; and performing feature enhancement operation on the soil moisture resolution-enhanced feature map to obtain the enhanced soil moisture data of the target area.

[0009] In some embodiments, the determining the soil moisture reconstruction feature map according to the assimilated soil moisture feature map and the microwave soil moisture continuous feature map comprises: correcting the assimilated soil moisture feature map according to the microwave soil moisture continuous feature map in a channel attention module to obtain an assimilated soil moisture corrected feature map; performing spatial information enhancement on the microwave soil moisture continuous feature map according to the assimilated soil moisture feature map in a spatial attention module to obtain a microwave soil moisture spatial enhanced feature map; and merging the microwave soil moisture spatial enhanced feature map and the assimilated soil moisture corrected feature map through weighted summation operation to obtain the soil moisture reconstruction feature map.

[0010] In some embodiments, the steps are implemented in a trained neural network model, and the neural network model comprises a missing reconstruction module, a multi-source fusion module, and a downscaling module; and a step-by-step training strategy is used to control the balance between the missing reconstruction module, the multi-source fusion module, and the downscaling module when training the neural network model.

[0011] In a second aspect, the present application provides an enhanced soil moisture data device, comprising: a first acquisition module configured to acquire satellite remote sensing soil moisture data, land surface assimilation soil moisture data, and environmental data of a target region within a certain time range; a first determination module configured to determine a soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within the certain time range; a second determination module configured to determine a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map; and a third determination module configured to determine enhanced soil moisture data with enhanced resolution according to the soil moisture multi-source fusion feature map and the environmental data.

[0012] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of any one of the first aspect.

[0013] In a fourth aspect, the present application provides a readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the method of any one of the first aspect.

[0014] In a fifth aspect, the present application provides a computer program product, comprising a computer program / instruction, wherein the computer program is executed by a processor to implement the steps of the method of any one of the first aspect.

[0015] The present application has the following beneficial effects: the present application acquires satellite remote sensing soil moisture data, land surface assimilation soil moisture data, and environmental data of a target region within a certain time range; determines a soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within the certain time range; determines a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map; and determines enhanced soil moisture data with enhanced resolution according to the soil moisture multi-source fusion feature map and the environmental data. The obtained enhanced soil moisture data has high consistency with site measured soil moisture, and can provide support of spatially and temporally continuous and high-resolution soil moisture data for related hydro-meteorological research. BRIEF DESCRIPTION OF DRAWINGS

[0016] The scope of the present disclosure can be better understood by reading the following detailed description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0017] Figure 1 A whole flow chart of a soil moisture data enhancement method provided for an embodiment of the present application;

[0018] Figure 2 A flow chart of a neural network model for soil moisture data enhancement provided for an embodiment of the present application;

[0019] Figure 3 A structure block diagram of a soil moisture data enhancement device provided for an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the drawings, and the described embodiments should not be regarded as limiting the present application, and all other embodiments obtained by those skilled in the art without making creative efforts fall within the scope of protection of the present application.

[0021] In the following description, "some embodiments" are related to a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0022] If there are similar descriptions of "first\second\third" in the application file, the following explanations are added, in the following description, the terms "first\second\third" related to only distinguish similar objects, and do not represent a specific order of the objects, and it can be understood that "first\second\third" can be interchanged with a specific order or sequence as allowed, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

[0024] Embodiment 1:

[0025] Soil moisture retrieved from microwave radiometric brightness temperature has high accuracy, but it is discontinuous in time and space and has low spatial resolution, which cannot meet the needs of fine soil moisture monitoring. Deep learning is a large deep artificial neural network tool that has attracted much attention in recent years. Multi-level networks can deeply mine the multi-scale spatial features and potential correlations of various environmental parameters and are widely used to solve various problems in the field of hydrology. However, there is still a lack of research on using deep learning to retrieve high-resolution and time-space-continuous soil moisture.

[0026] In view of the problems in the prior art, such as Figure 1 As shown in the accompanying drawings, the present application provides a soil moisture data enhancement method, which is applied to an electronic device, which can be a server, a mobile terminal, a computer, a cloud platform, etc. The function realized by the device data processing provided in the embodiments of the present application can be realized by calling program code by the processor of the electronic device, wherein the program code can be saved in a computer storage medium, and the soil moisture data enhancement method comprises the following steps:

[0027] Step S1: acquiring satellite remote sensing soil moisture data, land surface assimilation soil moisture data and environmental data of a target region within a certain time range.

[0028] The relevant data of the target region are collected from multiple data sources, including satellite remote sensing soil moisture data, land surface assimilation soil moisture data and environmental data. The satellite remote sensing soil moisture data come from the soil moisture data observed by the soil moisture active and passive satellite (SMAP) within a certain time range. The land surface assimilation soil moisture data come from the soil moisture data of the China Meteorological Administration Land Data Assimilation System (CLDAS). The environmental data include the longitude, latitude, altitude, nighttime surface temperature, vegetation index and drought index of the target region.

[0029] Step S2: determining a soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within a certain time range.

[0030] Due to the existence of revisit time, seasonal ice period and other factors, the soil moisture data retrieved by passive microwave remote sensing has a serious missing problem, which greatly affects the time-space continuity of the data. In a single collection, the true situation of the soil moisture of the target region cannot be collected, so multiple time phases need to be collected to form satellite remote sensing soil moisture data of the target region at different times. Therefore, in order to obtain more complete microwave soil moisture data of the target region, satellite remote sensing soil moisture data within a certain time range need to be acquired.

[0031] Therefore, in some embodiments, the step S2 "determining the soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within a certain time range" comprises:

[0032] Step S21: extracting effective spatial features of each time phase from the satellite remote sensing soil moisture data through a partial convolution layer.

[0033] Due to the low spatio-temporal coverage of satellite remote sensing soil moisture data, directly using a convolution layer to extract effective spatial features is easily affected by missing areas, therefore, the present application first uses a partial convolution layer to process the original input data. The partial convolution layer can distinguish missing and non-missing areas by automatically updating the mask, thereby significantly improving the training efficiency and performance. The extraction expression is as follows:

[0034]

[0035] In the formula, S ( l m,n) S l (m, n) represents the input data of the lth layer at the spatial position (m, n); M represents the mask corresponding to the input data, the pixel value of which is 0 or 1, indicating the missing or non-missing point state, respectively; 1 is a matrix with all elements being 1 in the same shape as M; W and b are the convolution kernel and the bias term, respectively. Through the operation of this formula, the output value is completely dependent on the non-missing area. In addition, the partial convolution layer also needs to update the mask synchronously. If there is at least one non-missing input value in the convolution operation, the mask at the corresponding position is updated to non-missing. The partial convolution block fully considers the large number of non-missing areas in the input data, and can improve the effectiveness of feature extraction.

[0036] Step S22: obtaining the time phase information of each effective spatial feature by using a long short-term memory block;

[0037] Step S23: fusing the effective spatial features of each time phase by using the time phase information to obtain the microwave soil moisture continuous feature map of the target region.

[0038] Only using spatial correlation cannot solve the problem of large-scale missing in the input data. Therefore, the present application introduces a long short-term memory block to fill in the missing areas of each other through the adjacent effective spatial features of the time phase information. The long short-term memory network is the earliest proposed gating algorithm, and each cycle module controls the unit state through three gating units, i.e. the forget gate, the input gate and the output gate. After the partial convolution block, the present application uses a convolution-long short-term memory block to extract adjacent time phase information, further fill in the large-scale missing areas in the input data, and obtain the spatio-temporal continuous microwave soil moisture continuous feature map.

[0039] Step S24: determining an assimilated soil moisture feature map of the target region according to the land surface assimilated soil moisture data.

[0040] Since the land surface assimilated soil moisture data does not have a missing problem, the conventional technical means can be directly used to extract the assimilated soil moisture feature map in the present application.

[0041] Step S25: determining the soil moisture reconstruction feature map according to the assimilated soil moisture feature map and the microwave soil moisture continuous feature map.

[0042] In some embodiments, step S25 "determining the soil moisture reconstruction feature map according to the assimilated soil moisture feature map and the microwave soil moisture continuous feature map" comprises:

[0043] Step S251: correcting the assimilated soil moisture feature map according to the microwave soil moisture continuous feature map in the channel attention module to obtain an assimilated soil moisture corrected feature map.

[0044] The assimilated soil moisture feature map has strong spatial continuity, but since it is obtained by data assimilation, there may be deviations in some positions. Therefore, the channel attention mechanism is used in the present application to combine the microwave soil moisture continuous feature map to correct the assimilated soil moisture feature map, thereby obtaining the assimilated soil moisture corrected feature map.

[0045] Step S252: the assimilated soil moisture feature map performs spatial information enhancement on the microwave soil moisture continuous feature map in the spatial attention module to obtain a microwave soil moisture spatial enhancement feature map.

[0046] Since the assimilated soil moisture feature map has good spatial continuity, and although the microwave soil moisture continuous feature map has been completed by the effective spatial features of adjacent time phases, the completed feature map still has a mismatch in spatial continuity. Therefore, the spatial attention mechanism is used to combine the assimilated soil moisture feature map to perform spatial enhancement on the microwave soil moisture continuous feature map, thereby obtaining the microwave soil moisture spatial enhancement feature map.

[0047] Step S253: merging the microwave soil moisture spatial enhancement feature map and the assimilated soil moisture corrected feature map by weighted summation operation to obtain the soil moisture reconstruction feature map.

[0048] Since the microwave soil moisture spatial enhancement feature map has good data accuracy, and the assimilated soil moisture corrected feature map has high spatial continuity, the fusion of the two can obtain a soil moisture reconstruction feature map with high spatial resolution and higher accuracy. The expression is as follows:

[0049]

[0050] S CroA = a x S SA + (1-a) x S CA

[0051] In the formula, S SA , S CA and S CroA respectively represent spatial attention, channel attention and output feature map after weighted summation operation; σ(·) represents Sigmoid function; W and b respectively represent convolution kernel and bias term; and are element multiplication operation and convolution operation respectively; a is a weight determined by the non-missing rate of homologous space-time information.

[0052] Step S3: determining a soil moisture multi-source fusion feature map according to the land surface assimilated soil moisture data and the soil moisture reconstructed feature map.

[0053] Step S2 only reconstructs the missing information of the soil moisture feature map, but to obtain a soil moisture data that is comprehensively enhanced, the resolution of the soil moisture reconstructed feature map needs to be enhanced. Therefore, in some embodiments, step S3 "determining a soil moisture multi-source fusion feature map according to the land surface assimilated soil moisture data and the soil moisture reconstructed feature map" comprises:

[0054] Step S31: determining soil moisture multi-level features of the target area according to the land surface assimilated soil moisture data.

[0055] Step S32: determining soil moisture multi-scale features of the target area according to the soil moisture multi-level features.

[0056] Step S33: determining the soil moisture multi-source fusion feature map according to the soil moisture multi-scale features and the soil moisture reconstructed feature map.

[0057] Since satellite remote sensing soil moisture data and land surface assimilated soil moisture data have their own advantages in terms of temporal phase consistency, spatial distribution consistency and numerical accuracy, both kinds of data can reflect the actual situation of soil moisture to a certain extent, but they show significant differences. In order to explore the complex nonlinear correlation between the two kinds of data, first, the residual dense block is used to extract multi-level features from the land surface assimilated soil moisture data. Then, the Laplace attention is used to extract multi-scale features from the multi-level features, and then the extracted multi-scale features are fused with the soil moisture reconstructed feature map, so that the fused feature map is preliminarily enhanced in terms of accuracy and resolution.

[0058] The Laplacian attention is composed of a global mean pooling operation, a multi-scale dilated convolution layer, and a sigmoid operation, and its expression is as follows:

[0059]

[0060] wherein g d represents the result of the global mean pooling operation; h and w represent the height and width of the input feature map, respectively; S (m,n) represents the input feature at the spatial position (m, n); S Lap and S input represent the output feature map and the input feature map of the Laplacian attention, respectively; D3(·), D5(·), and D7(·) represent the convolution layers with dilation rates of 3, 5, and 7, respectively; and are element multiplication and convolution operations, respectively; and σ(·) represents the sigmoid function; W and b represent the convolution kernel and the bias term, respectively.

[0061] Step S4: determining resolution-enhanced enhanced soil moisture data according to the soil moisture multi-source fusion feature map and the environmental data.

[0062] In the present application, the enhanced soil moisture data of the target region is intended to be obtained, although the spatio-temporally continuous soil moisture data is obtained in step S1, the spatial resolution thereof is low and cannot meet the needs of fine monitoring applications. Since the obtained data is large-scale data, and fine soil moisture data of the target region is intended to be obtained, the multi-source fusion feature map needs to be down-scaled.

[0063] Therefore, in some embodiments, step S4 "determining resolution-enhanced enhanced soil moisture data according to the soil moisture multi-source fusion feature map and the environmental data" comprises:

[0064] Step S41: determining the environmental features of the target region according to the environmental data.

[0065] In order to further improve the resolution of the soil moisture multi-source fusion feature map in the target region, the environmental features of the target region need to be integrated into the soil moisture multi-source fusion feature map to enhance the detailed information of the soil moisture multi-source feature map in the target region. Therefore, the environmental features need to be extracted from the environmental data in the present application. The environmental data includes the longitude, latitude, altitude, nighttime surface temperature, vegetation index, and drought index of the target region. Then the features of each environmental data, i.e., the environmental features, are extracted one by one through conventional means.

[0066] Step S42: performing resolution enhancement on the soil moisture multi-source fusion feature map according to the environmental features to obtain a soil moisture resolution-enhanced feature map.

[0067] After obtaining the environmental features, the environmental features are input into the cross-attention module one by one, so that the cross-attention module recalibrates the soil moisture multi-source fusion feature map according to the environmental features one by one, excavates the potential correlation between each environmental data and the soil moisture, and effectively enhances the spatial details of the soil moisture with the aid of high-resolution auxiliary information.

[0068] Step S43: performing a feature enhancement operation on the soil moisture resolution enhancement feature map to obtain enhanced soil moisture data of the target region.

[0069] In the downscaling stage, a dense residual attention block is used to perform a feature enhancement operation on the soil moisture resolution enhancement feature map. The dense residual attention block, also known as a residual dense convolution block with embedded attention, is composed of a three-layer structure containing a residual dense block and a residual attention block, and has been proven to be able to effectively extract high-level information from the feature map and play an important role in the spatial downscaling process. The classic residual dense block has obvious advantages in information extraction and transmission, so it is introduced into the main module to fully extract detailed information from various auxiliary factors. In addition, considering that the features from multiple variables have extremely high complexity, the residual attention is used to recalibrate the feature map to obtain more accurate soil moisture information. Specifically, it can be expressed as:

[0070]

[0071] In the formula, S input and S RAB respectively represent the input and output feature maps of the residual attention block, F CA and F SA respectively represent the output feature maps of the channel attention and spatial attention sub-blocks in the dense residual attention block, W RAB and b RAB respectively represent the convolution kernel and the bias term of the residual attention block. In summary, the dense residual attention module extracts and recalibrates multi-level soil moisture information through alternating residual dense blocks and residual attention blocks, and further estimates high-resolution soil moisture data.

[0072] In summary, the present method can be roughly divided into three steps. The first step is missing reconstruction, from missing to spatio-temporal continuity; the second step is fusion, to preliminarily improve the spatial resolution from 0.36 degrees to 0.06 degrees, and the detailed information comes from the assimilation data of 0.06 degrees; the third step is downscaling, to further improve the spatial resolution from 0.06 degrees to 0.01 degrees, and the detailed information comes from the environmental data of 0.01 degrees. Through the above three steps, the final fine reconstruction result has high consistency with the site measured soil moisture, and can provide spatio-temporal continuous and high-resolution soil moisture data support for related hydro-meteorological research.

[0073] In some embodiments, the above steps can be implemented by a trained neural network model. As shown in the figure, the neural network model can be roughly divided into "reconstruction module - multi-source fusion module - downscaling module" according to functions. Figure 2

[0074] The reconstruction module is mainly used to determine the soil moisture reconstruction feature map of the target area according to the satellite remote sensing soil moisture data and the land surface assimilated soil moisture data in a certain time range. The reconstruction module includes a partial convolution block, a long short-term memory block, a first cross attention block and a first convolution layer.

[0075] First, the partial convolution block is used to extract the effective spatial features of each phase from the satellite remote sensing soil moisture data, and then the long short-term memory block is used to extract the phase information of the effective spatial features. The missing area is reconstructed by using the space-time information to obtain the microwave soil moisture continuous feature map of the target area. Then, the assimilated soil moisture feature map is obtained according to the land surface assimilated soil moisture data. After obtaining the land surface assimilated soil moisture feature map and the microwave soil moisture continuous feature map, the assimilated soil moisture feature map is corrected according to the microwave soil moisture continuous feature map in the channel attention module of the first cross attention block to obtain the corrected assimilated soil moisture feature map, and the microwave soil moisture continuous feature map is enhanced in the spatial information according to the assimilated soil moisture feature map in the spatial attention module to obtain the microwave soil moisture spatial enhancement feature map. Then, the microwave soil moisture spatial enhancement feature map and the corrected assimilated soil moisture feature map are combined by weighted summation operation, and then the soil moisture reconstruction feature map is obtained after the convolution layer. The scale of the soil moisture reconstruction feature map is 0.36°, and the scale of the satellite remote sensing soil moisture data is also 0.36°.

[0076] The missing reconstruction process of the present application can be represented as:

[0077]

[0078] In the formula, z represents a variable with a resolution of 0.36°, including a soil moisture reconstruction feature z reco , satellite remote sensing soil moisture data z orig and down-sampled land surface assimilated soil moisture data z CLDAS ; T represents the time window (previous T days) of the homologous reference data, which is set to 7 days in the present application; f1(·) represents the reconstruction network.

[0079] ​The multi-source fusion module is mainly used for determining a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map. The multi-source fusion module comprises a dense residual block, a Laplacian attention block, a first up-sampling block, a second cross-attention block and a second convolutional layer.

[0080] The dense residual block is used for determining a soil moisture multi-level feature of the target area according to the land surface assimilation soil moisture data. The Laplacian attention block is connected with the dense residual block and is used for determining a soil moisture multi-scale feature of the target area according to the soil moisture multi-level feature. The first up-sampling block is connected with the first convolutional layer and the second cross-attention block and is used for obtaining an up-sampled soil moisture reconstruction feature map. The second cross-attention block is used for recalibrating and fusing the soil moisture multi-scale feature and the soil moisture reconstruction feature map, and the soil moisture multi-source fusion feature map is obtained after the second convolutional layer. The scale of the soil moisture multi-source fusion feature map is 0.06°, and the scale of the land surface assimilation soil moisture data is also 0.06°.

[0081]

[0082] In the formula, y represents a 0.06° resolution variable, including a soil moisture multi-source fusion feature map y fuse and the land surface assimilation soil moisture data y CLDAS ; f2(·) represents a fusion network.

[0083] The down-scaling module comprises a second up-sampling block, a multi-factor cross-attention block, a dense residual attention block and a third convolutional layer.

[0084] One end of the second up-sampling block is connected with the second convolutional layer, and the other end is connected with the multi-factor cross-attention block, and is used for obtaining an up-sampled soil moisture multi-source fusion feature map. The multi-factor cross-attention block is used for determining an environmental feature of the target area according to the environmental data, and performing resolution enhancement on the soil moisture multi-source fusion feature map according to the environmental feature, to obtain a soil moisture resolution enhanced feature map. The dense residual attention block is used for performing feature enhancement operation on the soil moisture resolution enhanced feature map, and the enhanced soil moisture data is obtained after the third convolutional layer, wherein the scale of the enhanced soil moisture data is 0.01°.

[0085] The multi-source fusion result is subjected to spatial down-scaling, and the process can be represented as:

[0086]

[0087] In the formula, x represents a 0.01° resolution variable, including enhanced soil moisture data x downwherein six auxiliary factors are latitude (LAT), longitude (LON), elevation (DEM), night land surface temperature (LSTN), vegetation index (NDVI) and multi-band drought index (NMDI); f3(·) represents a downscaling network.

[0088] In training the neural network model, a step-by-step training strategy is adopted to control the balance among the missing reconstruction module, the multi-source fusion module and the downscaling module.

[0089] The network trained can be constrained by a multi-task loss function, and when the overall loss function converges, the trained network is used for fine reconstruction test of soil moisture, and finally high-resolution spatiotemporal continuous soil moisture data are generated.

[0090] The overall loss function is constructed by combining the missing reconstruction loss, the multi-source fusion loss and the spatial downscaling loss, and the expression is as follows:

[0091]

[0092] Although the multi-task learning framework integrates the functions of missing reconstruction, multi-source fusion and spatial downscaling, the three are in a hierarchical progressive relationship, and when the loss of the missing reconstruction module is reduced, the multi-source fusion module can be guided, and when the loss of the multi-source fusion module is reduced, the spatial downscaling module can be guided. Therefore, the step-by-step training strategy is adopted to control the balance among the missing reconstruction, the multi-source fusion and the spatial downscaling, so as to ensure that the model can progressively complete the three tasks. The step-by-step training strategy is mainly realized by setting three coefficients λ1, λ2 and λ3, that is, in the initial stage, a higher λ1 value is set to make the model training focus on the missing reconstruction module; in the middle stage, a higher λ2 value is set to train the multi-source fusion module based on the more accurate missing reconstruction result; and when the training exceeds a specified number of iterations, a higher λ3 value is set to improve the spatial downscaling module based on the more accurate multi-source fusion result, so that the spatiotemporal fine soil moisture is obtained.

[0093] The present application makes full use of multi-source spatiotemporal and environmental data to realize the continuity and resolution enhancement of soil moisture. The main advantages are as follows: 1) integrating various attention mechanisms and multi-level convolution modules to gradually enhance the continuity and resolution of soil moisture data by taking land surface assimilation soil moisture data as a bridge; 2) designing a multi-task loss function and a step-by-step training strategy to effectively guide the modeling process, so that the model has higher robustness.

[0094] Compared with the existing soil moisture data fine reconstruction techniques based on physical and theoretical scale change decomposition, random forest, etc., the model can fully utilize the unique advantages of satellite remote sensing soil moisture data and land surface assimilation soil moisture data, introduce various auxiliary information, and realize high-resolution and spatial-temporal continuous soil moisture estimation. The fine reconstruction result has high consistency with the measured soil moisture at the site, and can provide spatial-temporal continuous and high-resolution soil moisture data support for related hydrological and meteorological researches.

[0095] Embodiment 2

[0096] Based on the foregoing embodiments, the embodiments of the present application provide an acceleration reduction device for a sample. Each module included in the device and each unit included in each module can be implemented by a processor in a computer device. Of course, it can also be implemented by a specific logic circuit. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0097] As shown in Figure 3 An enhanced device for soil moisture data includes a first acquisition module 1, a first determination module 2, a second determination module 3, and a third determination module 4.

[0098] The first acquisition module 1 is configured to acquire satellite remote sensing soil moisture data, land surface assimilation soil moisture data, and environmental data of a target region within a certain time range. The first determination module 2 is configured to determine a soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within the certain time range. The second determination module 3 is configured to determine a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map. The third determination module 4 is configured to determine enhanced soil moisture data with enhanced resolution according to the soil moisture multi-source fusion feature map and the environmental data.

[0099] Each module in the above-mentioned enhanced device for soil moisture data can be implemented by software, hardware, and a combination thereof, in whole or in part. Each module can be embedded in or independent of a processor in a device in hardware form, or can be stored in a memory in a processing device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module. It should be noted that the division of modules in the embodiments of the present application is illustrative, and is only a logical function division. In actual implementation, there can be another division manner.

[0100] Embodiment 3:

[0101] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of the first aspect.

[0102] Embodiment 4:

[0103] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program, wherein the computer program is executed by a processor to implement the steps of the method according to any one of the first aspect.

[0104] Embodiment 5:

[0105] In a fifth aspect, a computer program product is provided, comprising computer programs / instructions, wherein the computer program is executed by a processor to implement the steps of the method according to any one of the first aspect.

[0106] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0107] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0108] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0109] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0110] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. They can be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0111] Furthermore, in the various embodiments of this application, all functional units can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0112] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, and the foregoing program can be stored in a computer readable storage medium, and the program performs the steps of the above-mentioned method embodiments when executed. The foregoing storage medium includes a mobile storage device, a read only memory (ROM), a magnetic disc or an optical disc, and various storage medium capable of storing program codes.

[0113] Alternatively, the integrated units of the present application can be stored in a computer readable storage medium if they are realized in the form of software function modules and sold or used as independent products. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of software products, and the computer software products are stored in a storage medium, including a plurality of instructions for causing a controller to execute all or part of the methods described in the embodiments of the present application. The foregoing storage medium includes a mobile storage device, a ROM, a magnetic disc or an optical disc, and various storage medium capable of storing program codes.

[0114] The above is only an embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of augmenting soil moisture data, the method comprising: The method comprises the following steps: acquiring satellite remote sensing soil moisture data, land surface assimilation soil moisture data and environmental data of a target area within a certain time range; determining a soil moisture reconstruction feature map of the target area according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within the certain time range; the step of determining the soil moisture reconstruction feature map of the target area according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within the certain time range comprises: extracting effective spatial features of each phase from the satellite remote sensing soil moisture data through a partial convolution layer; acquiring phase information of each effective spatial feature; fusing the effective spatial features of each phase according to the phase information to obtain a microwave soil moisture continuous feature map of the target area; determining an assimilation soil moisture feature map of the target area according to the land surface assimilation soil moisture data; determining the soil moisture reconstruction feature map according to the assimilation soil moisture feature map and the microwave soil moisture continuous feature map; determining a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map; the step of determining the soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map comprises: determining a soil moisture multi-level feature of the target area according to the land surface assimilation soil moisture data; determining a soil moisture multi-scale feature of the target area according to the soil moisture multi-level feature; determining the soil moisture multi-source fusion feature map according to the soil moisture multi-scale feature and the soil moisture reconstruction feature map; determining enhanced soil moisture data with resolution enhancement according to the soil moisture multi-source fusion feature map and the environmental data; the step of determining the enhanced soil moisture data with resolution enhancement according to the soil moisture multi-source fusion feature map and the environmental data comprises: determining an environmental feature of the target area according to the environmental data; performing resolution enhancement on the soil moisture multi-source fusion feature map according to the environmental feature to obtain a soil moisture resolution enhancement feature map; performing feature enhancement operation on the soil moisture resolution enhancement feature map to obtain enhanced soil moisture data of the target area.

2. The method of claim 1, wherein, the step of determining the soil moisture reconstruction feature map according to the assimilation soil moisture feature map and the microwave soil moisture continuous feature map comprises: correcting the assimilation soil moisture feature map according to the microwave soil moisture continuous feature map in a channel attention module to obtain an assimilation soil moisture correction feature map; performing spatial information enhancement on the microwave soil moisture continuous feature map according to the assimilation soil moisture feature map in a spatial attention module to obtain a microwave soil moisture spatial enhancement feature map; merging the microwave soil moisture spatial enhancement feature map and the assimilation soil moisture correction feature map through weighted summation operation to obtain the soil moisture reconstruction feature map.

3. The method according to any of claims 1-2, characterized by, Further comprising: the above steps can be implemented in a trained neural network model; the neural network model comprises a missing reconstruction module, a multi-source fusion module and a downscaling module; In the training of the neural network model, a step-by-step training strategy is adopted to control the balance among the missing reconstruction module, the multi-source fusion module and the downscaling module.

4. An apparatus for enhancing soil moisture data, characterized by, Suitable for the method as claimed in any of claims 1-2, comprising: The first acquisition module is configured to acquire satellite remote sensing soil moisture data, land surface assimilation soil moisture data and environmental data of a target region within a certain time range. The first determination module is configured to determine a soil moisture reconstruction feature map of the target region according to the satellite remote sensing soil moisture data and the land surface assimilation soil moisture data within a certain time range. The second determination module is configured to determine a soil moisture multi-source fusion feature map according to the land surface assimilation soil moisture data and the soil moisture reconstruction feature map. The third determination module is configured to determine resolution-enhanced enhanced soil moisture data according to the soil moisture multi-source fusion feature map and the environmental data.

5. An electronic device, comprising: A computer program product comprising a memory, a processor and a computer program stored on the memory, the processor executing the computer program to implement the steps of the method of any one of claims 1 to 3.

6. A readable storage medium, having stored thereon a computer program, characterized in that, The computer program product, when executed by the processor, implements the steps of the method of any one of claims 1 to 3.

7. A computer program product comprising computer programs / instructions, characterized in that, The computer program product, when executed by the processor, implements the steps of the method of any one of claims 1 to 3. The computer program product, when executed by the processor, implements the steps of the method of any one of claims 1 to 3.

Images