Land surface temperature prediction method and system based on multi-layer neural network

Abstract

The application discloses a kind of based on multilayer neural network ground surface temperature prediction method and system, wherein the method includes: obtaining remote sensing image and its corresponding atmospheric external force data and ground surface temperature data;Remote sensing image is handled to obtain multiple different scale ground surface feature data, and NDVI data and NDBI data of corresponding scale are added to obtain multiple complete ground surface feature data;Ground surface temperature data is handled to obtain ground surface feature data corresponding ground surface temperature data;Ground surface temperature prediction model is constructed, and complete ground surface feature data, ground surface temperature data and atmospheric external force data are used as training set to carry out model training, to obtain trained ground surface temperature prediction model, to carry out ground surface temperature prediction according to trained ground surface temperature prediction model;Therefore, by multilayer perception neural network, since physical information constraint is applied, mapping relationship between input and output is learned from multiple scales, and good prediction effect can be achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing, and particularly relates to a multi-layer neural network land surface temperature prediction method, a computer readable storage medium, a computer device and a multi-layer neural network land surface temperature prediction system. BACKGROUND

[0002] In the related art, there are two main types of existing urban land surface temperature prediction methods. One type is a physics-based method, such as a physics-based model that uses the dynamic process between the urban canopy and the atmospheric boundary layer to solve the temperature field. However, this type of method is limited by the physical process used by the model, the accuracy of parameterization, and the availability of urban land surface feature datasets. In addition, it requires a huge amount of computing power and therefore has low accuracy. The other type is a statistical method, such as a method that establishes a relationship between land surface temperature and various observable data (such as meteorological variables, land cover, geography, and vegetation index) to solve the temperature. However, this type of method is limited by the complexity of the model, the availability and reliability of the data, the uncertainty of feature selection, and the lack of physical processes. It has poor generalization and low accuracy. SUMMARY

[0003] The present application aims to at least partially solve one of the above technical problems. To this end, one object of the present application is to propose a multi-layer neural network land surface temperature prediction method. The method uses a multi-layer perception neural network for prediction. Since physical information constraints are applied, the method can learn the mapping relationship between input and output from multiple scales, and can achieve good prediction results.

[0004] A second object of the present application is to propose a computer readable storage medium.

[0005] A third object of the present application is to propose a computer device.

[0006] A fourth object of the present application is to propose a multi-layer neural network land surface temperature prediction system.

[0007] To achieve the above object, the first aspect of the present application provides a land surface temperature prediction method based on a multi-layer neural network, comprising the following steps: obtaining remote sensing images and corresponding atmospheric external force data and land surface temperature data; processing the remote sensing images to obtain a plurality of different scale land surface feature data, and adding corresponding scale NDVI data and NDBI data to each different scale land surface feature data to obtain a plurality of different scale complete land surface feature data; decomposing and processing the land surface temperature data by using two-dimensional empirical mode decomposition to obtain land surface temperature data corresponding to each different scale land surface feature data; constructing a land surface temperature prediction model, and inputting the plurality of different scale complete land surface feature data, the land surface temperature data corresponding to each different scale land surface feature data and the atmospheric external force data as training set data into the land surface temperature prediction model for training to obtain a trained land surface temperature prediction model; obtaining a plurality of different scale complete land surface feature data and atmospheric external force data corresponding to a temperature area to be predicted, and inputting the same into the trained land surface temperature prediction model to obtain the land surface temperature of the temperature area to be predicted.

[0008] According to the land surface temperature prediction method based on a multi-layer neural network, first, remote sensing images and corresponding atmospheric external force data and land surface temperature data are obtained; then, the remote sensing images are processed to obtain a plurality of different scale land surface feature data, and corresponding scale NDVI data and NDBI data are added to each different scale land surface feature data to obtain a plurality of different scale complete land surface feature data; then, the land surface temperature data is decomposed and processed by using two-dimensional empirical mode decomposition to obtain land surface temperature data corresponding to each different scale land surface feature data; then, a land surface temperature prediction model is constructed, and the plurality of different scale complete land surface feature data, the land surface temperature data corresponding to each different scale land surface feature data and the atmospheric external force data are inputted as training set data into the land surface temperature prediction model for training to obtain a trained land surface temperature prediction model; finally, a plurality of different scale complete land surface feature data and atmospheric external force data corresponding to a temperature area to be predicted are obtained, and the same are inputted into the trained land surface temperature prediction model to obtain the land surface temperature of the temperature area to be predicted; thus, the mapping relationship between the input and the output is learned from multiple scales by the multi-layer perception neural network, and a better prediction effect can be achieved due to the application of physical information constraints.

[0009] In addition, the land surface temperature prediction method based on a multi-layer neural network according to the above embodiments of the present application can have the following additional technical features:

[0010] Optionally, the remote sensing image is processed to obtain a plurality of different scale ground surface feature data, comprising: predefining a plurality of geographic semantic labels; marking the remote sensing image according to the plurality of geographic semantic labels by using a geographic object marking network; constructing a plurality of geographic marking histograms according to the marked remote sensing image; and taking a to-be-predicted point as a center, a plurality of matrices composed of different units are constructed for each geographic marking histogram as a unit to obtain a plurality of different scale ground surface feature data.

[0011] Optionally, the plurality of geographic marking histograms are constructed by using the following formula:

[0012]

[0013] wherein S represents a category set on a specific unit, C(S) represents the number of points in S, S l represents the number of pixels of category I in S, and d(I) represents a geographic marking histogram of category I.

[0014] Optionally, the plurality of different scale complete ground surface feature data, the ground surface temperature data corresponding to each different scale ground surface feature data, and the atmospheric external force data are input into the ground surface temperature prediction model as training set data to train the trained ground surface temperature prediction model, comprising: encoding the plurality of different scale complete ground surface feature data to obtain corresponding ground surface encoded features; encoding the atmospheric external force data to obtain corresponding atmospheric encoded features; splicing the plurality of different scale ground surface encoded features and the corresponding atmospheric encoded features one by one to obtain a plurality of different scale final features; decoding the plurality of different scale final features to obtain corresponding ground surface temperature prediction results, so as to take the ground surface temperature prediction results and the ground surface temperature data as loss values; determining whether the trained ground surface temperature prediction model converges according to the loss values, so as to complete the training after determining that the trained ground surface temperature prediction model converges.

[0015] To achieve the above object, the second aspect of the present application provides a computer readable storage medium having a ground surface temperature prediction program based on a multi-layer neural network stored thereon, which realizes the ground surface temperature prediction method based on a multi-layer neural network when executed by a processor.

[0016] To achieve the above object, the third aspect of the present application provides a computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to realize the ground surface temperature prediction method based on a multi-layer neural network.

[0017] To achieve the above object, the fourth aspect of the present application provides a land surface temperature prediction system based on a multi-layer neural network, comprising: an acquisition module configured to acquire remote sensing images and corresponding atmospheric external force data and land surface temperature data; a processing module configured to process the remote sensing images to obtain a plurality of land surface feature data of different scales, and add NDVI data and NDBI data of corresponding scales on each land surface feature data of different scales to obtain a plurality of complete land surface feature data of different scales; a decomposition module configured to decompose and process the land surface temperature data by using two-dimensional empirical mode decomposition to obtain land surface temperature data corresponding to each land surface feature data of different scales; a model training module configured to construct a land surface temperature prediction model, and input the plurality of complete land surface feature data of different scales, the land surface temperature data corresponding to each land surface feature data of different scales and the atmospheric external force data as training set data into the land surface temperature prediction model for training to obtain a trained land surface temperature prediction model; and a prediction module configured to acquire a plurality of complete land surface feature data of different scales and atmospheric external force data corresponding to a temperature area to be predicted, and input the same into the trained land surface temperature prediction model to obtain the land surface temperature of the temperature area to be predicted.

[0018] According to the land surface temperature prediction system based on a multi-layer neural network, the acquisition module is configured to acquire remote sensing images and corresponding atmospheric external force data and land surface temperature data; the processing module is configured to process the remote sensing images to obtain a plurality of land surface feature data of different scales, and add NDVI data and NDBI data of corresponding scales on each land surface feature data of different scales to obtain a plurality of complete land surface feature data of different scales; the decomposition module is configured to decompose and process the land surface temperature data by using two-dimensional empirical mode decomposition to obtain land surface temperature data corresponding to each land surface feature data of different scales; the model training module is configured to construct a land surface temperature prediction model, and input the plurality of complete land surface feature data of different scales, the land surface temperature data corresponding to each land surface feature data of different scales and the atmospheric external force data as training set data into the land surface temperature prediction model for training to obtain a trained land surface temperature prediction model; and the prediction module is configured to acquire a plurality of complete land surface feature data of different scales and atmospheric external force data corresponding to a temperature area to be predicted, and input the same into the trained land surface temperature prediction model to obtain the land surface temperature of the temperature area to be predicted. Thus, the multi-layer perception neural network is used for prediction, the mapping relationship between the input and the output is learned from multiple scales due to the application of physical information constraints, and good prediction effect can be achieved.

[0019] In addition, the land surface temperature prediction system based on a multi-layer neural network according to the above-mentioned embodiments of the present application can have the following additional technical features:

[0020] Optionally, the processing module is further configured to: predefine a plurality of geographic semantic labels; mark the remote sensing image according to the plurality of geographic semantic labels by using a geographic object marking network; construct a plurality of geographic marking histograms according to the marked remote sensing image; and construct a plurality of matrices composed of different units according to the to-be-predicted point as a center and each geographic marking histogram as a unit, so as to obtain a plurality of ground surface feature data of different scales.

[0021] Optionally, the plurality of geographic marking histograms are constructed by using the following formula:

[0022]

[0023] wherein S represents a category set on a specific unit, C(S) represents the number of points in S, S l represents the number of pixels of category I in S, and d(I) represents the geographic marking histogram of category I.

[0024] Optionally, the model training module is further configured to: encode the plurality of ground surface feature data of different scales to obtain corresponding ground surface coded features; encode the atmospheric external force data to obtain corresponding atmospheric coded features; splice the plurality of ground surface coded features of different scales and the corresponding atmospheric coded features one by one to obtain a plurality of final features of different scales; decode the plurality of final features of different scales to obtain a corresponding ground surface temperature prediction result, so as to take the ground surface temperature prediction result and the ground surface temperature data as a loss value; determine whether the trained ground surface temperature prediction model converges according to the loss value, so as to complete the training after determining that the trained ground surface temperature prediction model converges. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 FIG. 1 is a flowchart of a ground surface temperature prediction method based on a multi-layer neural network according to an embodiment of the present application;

[0026] Figure 2 FIG. 4 is a network structure diagram of a ground surface temperature prediction model according to an embodiment of the present application;

[0027] Figure 3 FIG. 6 is a diagram of a ground surface temperature prediction result according to an embodiment of the present application;

[0028] Figure 4 FIG. 7 is a block diagram of a ground surface temperature prediction system based on a multi-layer neural network according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] Embodiments of the present application are described below in the following detailed description and in conjunction with the following figures, wherein like numbers refer to like elements throughout. The embodiments described below are merely exemplary and are not intended to limit the present application, except as provided in the appended claims.

[0030] For a better understanding of the above technical solutions, the exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0031] For a better understanding of the above technical solutions, the exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0032] As shown in the figure, the ground temperature prediction method based on the multi-layer neural network includes the following steps: Figure 1

[0033] S101, obtaining remote sensing images and corresponding atmospheric external force data and ground temperature data.

[0034] It should be noted that the above data can be directly downloaded through the network.

[0035] S102, processing the remote sensing images to obtain a plurality of ground feature data of different scales, and adding corresponding scale NDVI data and NDBI data on each ground feature data of different scales to obtain a plurality of complete ground feature data of different scales.

[0036] As an embodiment, processing the remote sensing images to obtain a plurality of ground feature data of different scales includes:

[0037] Predefining a plurality of geographic semantic labels; marking the remote sensing images according to the plurality of geographic semantic labels using a geographic feature marking network; constructing a plurality of geographic marking histograms according to the marked remote sensing images; taking a point to be predicted as the center, and establishing a plurality of matrices composed of different units for each geographic marking histogram as a unit to obtain a plurality of ground feature data of different scales.

[0038] As an embodiment, the following formula is used to construct a plurality of geographic marking histograms:

[0039]

[0040] wherein S represents a class set on a specific unit, C(S) represents the number of points in S, l represents the number of units in each geographic marking histogram, and H represents the number of geographic marking histograms.​l d(l) represents the number of pixels of category l in S, and d(l) represents the geotag histogram of category l.

[0041] It should be noted that if topographic features are extracted directly from remote sensing images, multilayer neural networks will find it difficult to extract information related to surface temperature prediction from complex graphic features; however, surface features at multiple different scales have clear structures and contain rich surface structure information, which can help the network better perceive and capture the relationship between surface structure and temperature.

[0042] As a specific example:

[0043] First, collect remote sensing images with a resolution of 1 meter corresponding to key cities across the country;

[0044] Then, multiple geographic semantic labels are predefined, such as water, buildings, vegetation, soil, roads and others, so that the remote sensing images of key cities collected can be labeled using the feature labeling network based on the multiple geographic semantic labels;

[0045] Next, a histogram of ground feature markers with a precision of 5m is constructed using the following formula:

[0046]

[0047] Where S represents the set of categories on a specific unit, and C(*) represents the number of points in S. l This represents the number of pixels of category l in S. The final feature marker histogram d(l) is a 6*1 vector.

[0048] Next, for a 60*60 area, a larger matrix is ​​built around this area using a batch of 5×5m cells; the matrix size is M*M*d, where M represents the number of cells and d represents the size of the feature vector of a cell, so as to construct multi-scale surface feature data (Geographical Semantic Category Fraction Matrix, GSCFM) according to different M values, where M can be 35, 45 and 55 respectively, that is, the cut size is 35×35, 45×45, 55×55.

[0049] Finally, NDVI (Normalized Difference Vegetation Index) and NDBI (Normalized Difference Built-up Index) data of corresponding sizes are added to the surface feature data for each multi-scale dataset.

[0050] S103, decompose the land surface temperature data by using two-dimensional empirical mode decomposition to obtain land surface temperature data corresponding to each different scale land surface feature data.

[0051] As an embodiment, the precision of the land surface temperature data can be 60 m, and the land surface temperature data is decomposed by using two-dimensional empirical mode decomposition (BEMD). Empirical mode decomposition (EMD) is a classic signal analysis tool that decomposes data into a series of low-frequency bases called "intrinsic mode functions (IMF)". The BEMD algorithm treats images as signals and decomposes them to obtain different IMFs, each containing information at different scales. Therefore, the connection between the IMFs of the land surface temperature image and the GSCFM at different scales is established, enabling the network to capture the relationship between urban land surfaces and temperature at different scales, thereby obtaining better estimation results. Considering that the land surface temperature data is used as a single-channel image for decomposition by BEMD, the formula is as follows:

[0052]

[0053] where n represents the number of iterations, h(*) is the intrinsic mode representing fine-scale distribution information. r(*) is the residual layer obtained by averaging the envelopes of local maxima and minima, and p represents the land surface temperature image.

[0054] S104, constructing a land surface temperature prediction model, and inputting the multiple different scale complete land surface feature data, the land surface temperature data corresponding to each different scale land surface feature data, and the atmospheric external force data as training set data into the land surface temperature prediction model for training to obtain a trained land surface temperature prediction model.

[0055] As an embodiment, the multiple different scale complete land surface feature data, the land surface temperature data corresponding to each different scale land surface feature data, and the atmospheric external force data are inputted as training set data into the land surface temperature prediction model for training to obtain a trained land surface temperature prediction model, including: encoding the multiple different scale complete land surface feature data to obtain corresponding land coding features; encoding the atmospheric external force data to obtain corresponding atmospheric coding features; one-by-one splicing the multiple different scale land coding features and the corresponding atmospheric coding features to obtain multiple different scale final features; decoding the multiple different scale final features to obtain corresponding land surface temperature prediction results, so as to take the land surface temperature prediction results and the land surface temperature data as loss values; determining whether the trained land surface temperature prediction model converges according to the loss values, so as to complete the training after determining that the trained land surface temperature prediction model converges.

[0056] As a specific embodiment, as shown in Figure 2 The ground surface temperature prediction model is a multi-layer neural network, and the structure thereof comprises a multi-scale encoder and an information parallel decoder; a residual network is used to encode complete ground surface feature data of multiple different scales to obtain ground surface encoded features F; a multi-layer neural network (MLP) is used to encode main atmospheric forcing variables in a physics-based climate model in consideration of the influence of atmospheric state on the ground surface temperature (LST) to obtain atmospheric encoded features G; the ground surface encoded features F of different scales are connected with corresponding atmospheric encoded features G to obtain multiple final features H of a corresponding 60*60m area; one scale corresponds to one branch, and the multiple final features H are input into an MLP layer for decoding, corresponding to the BEMD decomposition result, and using RMSE (Root Mean Square Error) as a loss function, back propagation is performed to adjust overall network parameters; the loss function formula is as follows:

[0057]

[0058] Wherein, n represents the number of samples, y represents the true value, and z represents the predicted value.

[0059] S105, obtaining multiple different scale complete ground surface feature data and atmospheric external force data corresponding to the temperature area to be predicted, and inputting the same into the trained ground surface temperature prediction model to obtain the ground surface temperature of the temperature area to be predicted.

[0060] That is, after the ground surface temperature prediction model is trained, the GSCFM data, NDVI data, NDBI data and atmospheric external force data corresponding to the area to be predicted are directly obtained and input into the ground surface temperature prediction model, so that the corresponding ground surface temperature can be obtained.

[0061] Figure 3 A schematic diagram of the ground surface temperature prediction result is shown in Figure 3 The temperature of cities such as Yinchuan, Shenyang and Chongqing in China is predicted, and a corresponding prediction result schematic diagram is obtained; wherein the numbers of each color bar represent the temperature, and the unit is Celsius; the vertical coordinate represents the city name, and the horizontal coordinate represents the true value, the predicted value and the error, respectively.

[0062] In summary, according to the method for predicting ground surface temperature based on a multi-layer neural network, first, remote sensing images and corresponding atmospheric external force data and ground surface temperature data are acquired; then, the remote sensing images are processed to obtain a plurality of ground surface feature data of different scales, and NDVI data and NDBI data of corresponding scales are added to each ground surface feature data of different scales to obtain a plurality of complete ground surface feature data of different scales; then, the ground surface temperature data are decomposed by using two-dimensional empirical mode decomposition to obtain ground surface temperature data corresponding to each ground surface feature data of different scales; then, a ground surface temperature prediction model is constructed, and the plurality of complete ground surface feature data of different scales, the ground surface temperature data corresponding to each ground surface feature data of different scales and the atmospheric external force data are input into the ground surface temperature prediction model as training set data to train the ground surface temperature prediction model to obtain a trained ground surface temperature prediction model; finally, the plurality of complete ground surface feature data of different scales and the atmospheric external force data corresponding to a temperature area to be predicted are acquired and input into the trained ground surface temperature prediction model to obtain the ground surface temperature of the temperature area to be predicted. Thus, the multi-layer perception neural network can effectively improve the temperature prediction effect, the atmospheric state physical quantity is introduced to increase the network interpretability, and the ground surface temperature can be more accurately predicted when new ground surface remote sensing images and atmospheric states are available, thereby providing a new effective method for urban micro-environment simulation technology.

[0063] In addition, the embodiment of the present application further provides a computer readable storage medium, which stores a ground surface temperature prediction program based on a multi-layer neural network, and the ground surface temperature prediction program based on the multi-layer neural network realizes the ground surface temperature prediction method based on the multi-layer neural network when executed by a processor.

[0064] The computer readable storage medium according to the embodiment of the present application realizes the ground surface temperature prediction method based on the multi-layer neural network through the ground surface temperature prediction program based on the multi-layer neural network, and the ground surface temperature prediction method based on the multi-layer neural network is realized when the ground surface temperature prediction program based on the multi-layer neural network is executed by the processor. Thus, the multi-layer perception neural network is used for prediction, the mapping relationship between the input and the output is learned from multiple scales due to the application of physical information constraints, and a better prediction effect can be achieved.

[0065] In addition, the embodiment of the present application further provides a computer device, which includes a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor realizes the ground surface temperature prediction method based on the multi-layer neural network when executing the program.

[0066] The computer device according to the embodiment of the present application stores the land surface temperature prediction program based on the multi-layer neural network in the memory, so that the land surface temperature prediction program based on the multi-layer neural network is executed by the processor to implement the land surface temperature prediction method based on the multi-layer neural network. Thus, the multi-layer perception neural network is used for prediction, the mapping relationship between the input and the output is learned from multiple scales due to the physical information constraint, and good prediction effect can be achieved.

[0067] In order to implement the above-mentioned embodiments, the embodiment of the present application proposes a land surface temperature prediction system based on a multi-layer neural network, as shown in the figure. Figure 4 The land surface temperature prediction system based on the multi-layer neural network comprises an acquisition module 10, a processing module 20, a decomposition module 30, a model training module 40 and a prediction module 50.

[0068] The acquisition module 10 is configured to acquire remote sensing images and corresponding atmospheric external force data and land surface temperature data. The processing module 20 is configured to process the remote sensing images to obtain a plurality of land surface feature data of different scales, and add NDVI data and NDBI data of corresponding scales on each land surface feature data of different scales to obtain a plurality of complete land surface feature data of different scales. The decomposition module 30 is configured to decompose and process the land surface temperature data by using two-dimensional empirical mode decomposition to obtain land surface temperature data corresponding to each land surface feature data of different scales. The model training module 40 is configured to construct a land surface temperature prediction model, and input the plurality of complete land surface feature data of different scales, the land surface temperature data corresponding to each land surface feature data of different scales and the atmospheric external force data as training set data into the land surface temperature prediction model for training to obtain a trained land surface temperature prediction model. The prediction module 50 is configured to acquire a plurality of complete land surface feature data of different scales and atmospheric external force data corresponding to a temperature area to be predicted, and input the same into the trained land surface temperature prediction model to obtain the land surface temperature of the temperature area to be predicted.

[0069] In some embodiments, the processing module 20 is further configured to predefine a plurality of geographic semantic labels, mark the remote sensing images according to the plurality of geographic semantic labels by using a geographic feature marking network, construct a plurality of geographic marking histograms according to the marked remote sensing images, and establish a plurality of matrices composed of different units by taking a point to be predicted as a center and each geographic marking histogram as a unit to obtain a plurality of land surface feature data of different scales.

[0070] In some embodiments, the plurality of geographic marking histograms are constructed by using the following formula:

[0071]

[0072] wherein S represents a category set on a specific unit, C(S) represents the number of points in S, and Sl wherein n(S) denotes the number of pixels in S, and d(l) denotes the geotag histogram of class l.

[0073] In some embodiments, the model training module 40 is further configured to encode the complete ground surface feature data of the plurality of different scales to obtain corresponding ground surface encoded features, encode the atmospheric external force data to obtain corresponding atmospheric encoded features, concatenate the plurality of different scales of ground surface encoded features and corresponding atmospheric encoded features one by one to obtain a plurality of different scales of final features, decode the plurality of different scales of final features to obtain corresponding ground surface temperature prediction results, take the ground surface temperature prediction results and the ground surface temperature data as loss values, determine whether the trained ground surface temperature prediction model converges according to the loss values, and complete the training when it is determined that the trained ground surface temperature prediction model converges.

[0074] In summary, according to the ground surface temperature prediction system based on the multi-layer neural network, the remote sensing image, the corresponding atmospheric external force data and the ground surface temperature data are obtained by the acquisition module; the processing module processes the remote sensing image to obtain a plurality of different scales of ground surface feature data, and adds the NDVI data and the NDBI data of the corresponding scale to each different scale of ground surface feature data to obtain a plurality of different scales of complete ground surface feature data; the decomposition module decomposes and processes the ground surface temperature data by using the two-dimensional empirical mode decomposition to obtain the ground surface temperature data corresponding to each different scale of ground surface feature data; the model training module constructs a ground surface temperature prediction model, and inputs the plurality of different scales of complete ground surface feature data, the ground surface temperature data corresponding to each different scale of ground surface feature data and the atmospheric external force data as training set data into the ground surface temperature prediction model for training to obtain a trained ground surface temperature prediction model; and the prediction module obtains the plurality of different scales of complete ground surface feature data and the atmospheric external force data corresponding to the temperature region to be predicted, and inputs the same into the trained ground surface temperature prediction model to obtain the ground surface temperature of the temperature region to be predicted. Thus, the multi-layer perception neural network is used for prediction, the mapping relationship between the input and the output is learned from the multiple scales due to the physical information constraint, and a better prediction effect can be achieved.

[0075] Those skilled in the art should understand that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] The present application is described in reference to the flowchart and / or block diagrams of the methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more of the flowchart and / or block diagrams. Figure 1 one or more of the flowchart and / or block diagrams.

[0077] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more of the flowchart and / or block diagrams. Figure 1 one or more of the flowchart and / or block diagrams.

[0078] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more of the flowchart and / or block diagrams. Figure 1 one or more of the flowchart and / or block diagrams.

[0079] It should be noted that any references made in the claims to an "apparatus" or "means" should not be construed to cover the corresponding structures only. Rather, such phrases are intended to also cover the corresponding processes which operate to produce the desired result. It should further be noted that the terms "first", "second", "third", etc. are used herein for purposes of nomenclature only and are not intended to impose or imply any importance or order of magnitude. It should also be noted that the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0080] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to cover all such variations and modifications as fall within the scope of the application.

[0081] Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the spirit and scope of the application. Accordingly, it is intended that all such modifications and changes be included in the present application only insofar as they fall within the scope of the claims and their equivalents.

[0082] In the description of the present application, it is to be understood that the terms "first", "second", "third" and the like, merely mean different instances of the same thing, and do not imply relative importance or a number of the indicated technical features. Thus, a feature defined with "first", "second", "third" can include one or more of the features. In the description of the present application, the meaning of "plurality" is two or more, unless specifically defined otherwise.

[0083] In the present application, unless specifically defined and limited otherwise, the terms "mounting", "connected", "connecting", "fixed", and the like, should be understood broadly, for example, can be fixed connection, can be detachable connection, or integral; can be mechanical connection, can be electrical connection; can be directly connected, can be indirectly connected through intermediate medium, can be internal communication of two elements or interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0084] In the present application, unless specifically defined and limited otherwise, the first feature "on" or "under" the second feature can be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, the first feature "above", "over" and "on" the second feature can be that the first feature is directly above or obliquely above the second feature, or only means that the horizontal height of the first feature is higher than that of the second feature. The first feature "below", "under" and "under" the second feature can be that the first feature is directly below or obliquely below the second feature, or only means that the horizontal height of the first feature is less than that of the second feature.

[0085] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms should not be understood as necessarily referring to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in the present application and the features of different embodiments or examples without contradiction.

[0086] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application, and that changes, modifications, substitutions and variations can be made by those skilled in the art without departing from the scope of the present application.

Claims

1. A land surface temperature prediction method based on a multi-layer neural network, characterized by, The method comprises the following steps: obtaining remote sensing images and corresponding atmospheric external force data and ground temperature data; processing the remote sensing images to obtain a plurality of ground feature data of different scales, and adding NDVI data and NDBI data of corresponding scales to each ground feature data of different scales to obtain a plurality of complete ground feature data of different scales; performing decomposition processing on the ground temperature data by using two-dimensional empirical mode decomposition to obtain ground temperature data corresponding to each ground feature data of different scales; constructing a ground temperature prediction model, and inputting the plurality of complete ground feature data of different scales, the ground temperature data corresponding to each ground feature data of different scales, and the atmospheric external force data as training set data into the ground temperature prediction model for training to obtain a trained ground temperature prediction model; obtaining a plurality of complete ground feature data of different scales and atmospheric external force data corresponding to a to-be-predicted temperature region, and inputting the same into the trained ground temperature prediction model to obtain the ground temperature of the to-be-predicted temperature region; wherein the processing of the remote sensing images to obtain a plurality of ground feature data of different scales comprises: predefining a plurality of geographic semantic labels; labeling the remote sensing images according to the plurality of geographic semantic labels by using a geographic feature labeling network; constructing a plurality of geographic label histograms according to the labeled remote sensing images; centering on a to-be-predicted point, a plurality of matrices composed of different units are established for each geographic label histogram to obtain a plurality of ground feature data of different scales; wherein the plurality of geographic label histograms are constructed by using the following formula: where S denotes a set of classes on a particular unit, C(S) denotes the number of points in S l denotes the number of pixels in S of class / , and d(l) denotes a geo-tag histogram of class / . 2.The method of claim 1, wherein, inputting the plurality of complete ground feature data of different scales, the ground temperature data corresponding to each ground feature data of different scales, and the atmospheric external force data as training set data into the ground temperature prediction model for training to obtain a trained ground temperature prediction model, comprising: encoding the plurality of complete ground feature data of different scales to obtain corresponding ground encoded features; encoding the atmospheric external force data to obtain corresponding atmospheric encoded features; splicing the ground encoded features of different scales and the corresponding atmospheric encoded features one by one to obtain a plurality of final features of different scales; decoding the plurality of final features of different scales to obtain corresponding ground temperature prediction results, so as to take the ground temperature prediction results and the ground temperature data as loss values; determining whether the trained ground temperature prediction model converges according to the loss values, so as to complete the training after determining that the trained ground temperature prediction model converges.

3. A computer-readable storage medium, characterized in that, The processor executes the program to implement the ground temperature prediction method based on a multi-layer neural network according to any one of claims 1-2.

4. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the ground temperature prediction method based on a multi-layer neural network according to any one of claims 1-2.

5. A land surface temperature prediction system based on a multi-layer neural network, characterized by, The program comprises: An acquisition module is configured to acquire remote sensing images, corresponding atmospheric external force data and ground surface temperature data; A processing module is configured to process the remote sensing images to obtain a plurality of ground surface feature data of different scales, and add corresponding scale NDVI data and NDBI data to each ground surface feature data of different scales to obtain a plurality of complete ground surface feature data of different scales; A decomposition module is configured to decompose the ground surface temperature data by using two-dimensional empirical mode decomposition to obtain corresponding ground surface temperature data of each ground surface feature data of different scales; A model training module is configured to construct a ground surface temperature prediction model, and input the plurality of complete ground surface feature data of different scales, the corresponding ground surface temperature data of each ground surface feature data of different scales and the atmospheric external force data as training set data into the ground surface temperature prediction model for training to obtain a trained ground surface temperature prediction model; A prediction module is configured to acquire a plurality of complete ground surface feature data of different scales and atmospheric external force data corresponding to a to-be-predicted temperature region, and input the same into the trained ground surface temperature prediction model to obtain the ground surface temperature of the to-be-predicted temperature region; The processing module is further configured to, predefine a plurality of geographic semantic labels; label the remote sensing images according to the plurality of geographic semantic labels by using a geographic feature labeling network; construct a plurality of geographic label histograms according to the labeled remote sensing images; establish a plurality of matrices composed of different units with the to-be-predicted point as the center and each geographic label histogram as a unit to obtain a plurality of ground surface feature data of different scales; The model training module is further configured to, where S denotes a set of classes on a particular unit, C(S) denotes the number of points in S l denotes the number of pixels in S of class / , and d(l) denotes a geo-tag histogram of class / .

6. The multi-layer neural network-based land surface temperature prediction system of claim 5, wherein, encode the plurality of complete ground surface feature data of different scales to obtain corresponding ground surface coded features; encode the atmospheric external force data to obtain corresponding atmospheric coded features; splicing the plurality of ground surface coded features of different scales and the corresponding atmospheric coded features one by one to obtain a plurality of final features of different scales; decode the plurality of final features of different scales to obtain corresponding ground surface temperature prediction results, so as to take the ground surface temperature prediction results and the ground surface temperature data as loss values; determine whether the trained ground surface temperature prediction model converges according to the loss values, so as to complete the training after determining that the trained ground surface temperature prediction model converges. ​

Images