Method for obtaining high-resolution DEM data and electronic equipment

Abstract

The invention provides a high-resolution DEM data obtaining method and electronic equipment, and the method comprises the steps: obtaining low-resolution DEM data; obtaining high-resolution DEM data according to the low-resolution DEM data in combination with a first generator; wherein the first generator is obtained by training the second generator according to the high-resolution DEM sample data set. According to the invention, the acquisition difficulty of the high-resolution DEM data can be reduced, and the acquisition precision of the high-resolution DEM data can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a method for obtaining high-resolution DEM data and electronic equipment. Background Art

[0002] High-resolution DEM (digital elevation model) data can accurately describe geographic information and is widely used in urban modeling, disaster assessment, hydrological modeling, military strategy, etc. However, in practical applications, it is difficult to obtain high-resolution DEM data and the accuracy is low. Summary of the invention

[0003] In order to solve the above problems, the present invention provides a method and electronic device for obtaining high-resolution DEM data, which can reduce the difficulty of obtaining high-resolution DEM data and improve the accuracy of obtaining high-resolution DEM data.

[0004] In a first aspect, the present invention provides a method for obtaining high-resolution DEM data, comprising: obtaining low-resolution DEM data; obtaining high-resolution DEM data based on the low-resolution DEM data and in combination with a first generator; wherein the first generator is obtained by training a second generator based on a high-resolution DEM sample data set.

[0005] Optionally, the method further includes: training the second generator according to the high-resolution DEM sample data set to obtain the first generator; wherein: the training of the second generator according to the high-resolution DEM sample data set includes: dividing the high-resolution DEM sample data set into multiple high-resolution DEM sample data groups; down-sampling the multiple high-resolution DEM sample data groups to obtain multiple low-resolution DEM sample data groups; wherein the multiple high-resolution DEM sample data groups correspond one-to-one to the multiple low-resolution DEM sample data groups; constructing a first training network; wherein the first training network includes the second generator, a discriminator and a first loss function; according to the first low-resolution DEM sample data group, training the second generator to obtain a first high-resolution DEM prediction data group; wherein the first low-resolution DEM sample data group is one of the multiple low-resolution DEM sample data groups; through the discriminator, the first high-resolution DEM prediction data group is trained. The method comprises the steps of: determining a first evaluation information based on a first high-resolution DEM sample data group and a first high-resolution DEM sample data group to obtain first discrimination information; wherein the first high-resolution DEM sample data group is a high-resolution DEM sample data group corresponding to the first low-resolution DEM sample data group in the multiple high-resolution DEM sample data groups; determining first evaluation information based on the first discrimination information and the first high-resolution DEM prediction data group, and combining the first loss function; adjusting the values ​​of parameters in the second generator and the discriminator based on the first evaluation information, and iteratively training the second generator, the discriminator and the first loss function after adjusting the values ​​of the parameters based on the second low-resolution DEM sample data group until the number of iterations is reached; wherein the second low-resolution DEM sample data group is one of the multiple low-resolution DEM sample data groups, and the second low-resolution DEM sample data group is different from the first low-resolution DEM sample data group; and determining the second generator obtained from the last iterative training as the first generator.

[0006] Optionally, before constructing the first training network, the method further includes: constructing a second training network; wherein the second training network includes a third generator and a second loss function; according to a third low-resolution DEM sample data group, training the third generator to obtain a third high-resolution DEM prediction data group; wherein the third low-resolution DEM sample data group is one of the multiple low-resolution DEM sample data groups, and the third low-resolution DEM sample data group is different from the second low-resolution DEM sample data group and the first low-resolution DEM sample data group; according to the third high-resolution DEM prediction data group and the third high-resolution DEM sample data group, and in combination with the second loss function, determining second evaluation information; wherein the third high-resolution DEM sample data group is a high-resolution DEM sample data group corresponding to the third low-resolution DEM sample data group from a plurality of high-resolution DEM sample data groups; adjusting the values ​​of the parameters in the third generator according to the second evaluation information, and iteratively training the third generator after adjusting the values ​​of the parameters and the second loss function based on a fourth low-resolution DEM sample data group until the number of iterations is reached; wherein the fourth low-resolution DEM sample data group is one of the plurality of low-resolution DEM sample data groups, and the fourth low-resolution DEM sample data group is different from the first low-resolution DEM sample data group, the second low-resolution DEM sample data group and the third low-resolution DEM sample data group; and determining the third generator obtained from the last iterative training as the second generator.

[0007] Optionally, the method also includes: after completing the last iterative training of the second generator, obtaining a high-resolution DEM verification dataset; downsampling the high-resolution DEM verification dataset to a low-resolution DEM verification dataset; determining a high-resolution DEM verification prediction set based on the low-resolution DEM verification dataset and in combination with the second generator obtained from the last iterative training; judging whether the second generator obtained from the last iterative training has passed the verification based on the high-resolution DEM verification prediction set and the high-resolution DEM verification dataset; if so, determining the second generator obtained from the last iterative training as the first generator; if not, adjusting the structure of the first training network.

[0008] Optionally, the second generator includes multiple deep residual dense modules, multiple adaptive residual convolution modules and sub-pixel convolution layers.

[0009] Optionally, each of the deep residual dense modules includes multiple dense residual modules, each of the dense residual modules includes multiple convolutional layers and a first activation function; wherein the multiple convolutional layers adopt dense connections; each of the adaptive residual convolution modules includes multiple variable convolutional layers and a second activation function; wherein the multiple variable convolutional layers adopt jump connections.

[0010] Optionally, the discriminator includes a U-Net framework, an attention mechanism gate module and a splicing module; the splicing module is used to realize information transfer between the U-Net framework and the attention mechanism gate module.

[0011] Optionally, the first evaluation information includes global evaluation information and adversarial evaluation information; and determining the first evaluation information according to the first discrimination information and the first high-resolution DEM prediction data group and in combination with the first loss function includes: determining the global evaluation information by the following formula:

[0012] L content =L mse +βL φmse

[0013]

[0014]

[0015] Among them, L content is the global evaluation information, L mse is the first global evaluation information, L φmse is the second global evaluation information, β is 0.006, N1 is the number of high-resolution DEM prediction data in the first high-resolution DEM prediction data group, 1≤i≤N1, I HR is a high-resolution DEM sample data in the first high-resolution DEM sample data group, I LR is a low-resolution DEM sample data in the first low-resolution DEM sample data group, For the second generator, is a high-resolution DEM prediction data in the first high-resolution DEM prediction data group, φ(I HR ) is I HR The characteristic data of for characteristic data;

[0016] The adversarial evaluation information is determined by the following formula:

[0017]

[0018] Among them, L Genis the adversarial evaluation information, N1 is the number of high-resolution DEM prediction data in the first high-resolution DEM prediction data group, 1≤i≤N1, D is the discriminator, It is a high-resolution DEM prediction data in the first high-resolution DEM prediction data group.

[0019] Optionally, determining the second evaluation information according to the third high-resolution DEM prediction data group and the third high-resolution DEM sample data group in combination with the second loss function includes: determining the second evaluation information by the following formula:

[0020]

[0021] Among them, L Gen is the second evaluation information, N2 is the number of high-resolution DEM prediction data in the third high-resolution DEM prediction data group, 1≤j≤N2, h j is the j-th high-resolution DEM prediction data in the third high-resolution DEM prediction data group, It is the j-th high-resolution DEM sample data in the third high-resolution DEM sample data group.

[0022] In a second aspect, the present application provides an electronic device comprising: a memory and a processor; the memory is used to store a computer program; the processor is used to implement the method for obtaining high-resolution DEM data described in any one of the first aspects when executing the computer program.

[0023] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for obtaining high-resolution DEM data as described in the first aspect is implemented.

[0024] The method for obtaining high-resolution DEM data of the present invention has the beneficial effect that: since the first generator is obtained by training the second generator according to the high-resolution DEM sample data set, the accuracy of the first generator is improved. In this way, high-resolution DEM data is obtained according to low-resolution DEM data and combined with the first generator, and the accuracy of obtaining high-resolution DEM data is improved. In addition, since high-resolution DEM data can be obtained based on the first generator, the difficulty and cost of obtaining high-resolution DEM data are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A schematic diagram of a process for obtaining high-resolution DEM data provided in an embodiment of the present application;

[0026] Figure 2A schematic diagram of the training process of the second generator provided in an embodiment of the present application;

[0027] Figure 3 A schematic diagram of the training process of the third generator provided in an embodiment of the present application;

[0028] Figure 4 A schematic diagram of the verification process of the second generator provided in an embodiment of the present application;

[0029] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be interpreted as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not intended to limit the scope of protection of the present invention.

[0031] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0032] The term "including" and its variations used in this article are open inclusions, that is, "including but not limited to"; the term "based on" means "based at least in part on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present invention are only used to distinguish between different devices, modules or units and other objects, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units and other objects.

[0033] It should be noted that the modifications of "one" and "plurality" mentioned in the present invention are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0034] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes, and are not used to limit the scope of these messages or information.

[0035] In order to solve the above technical problems, an embodiment of the present application provides a method for obtaining high-resolution DEM data, which can be executed by electronic devices, including but not limited to laptops, desktop computers, netbooks, tablet computers, digital readers, etc.

[0036] like Figure 1 As shown, the method for obtaining the high-resolution DEM data may include:

[0037] 101. Obtain low-resolution DEM data.

[0038] 102. According to the low-resolution DEM data and in combination with the first generator, high-resolution DEM data is obtained.

[0039] The first generator is obtained by training the second generator according to the high-resolution DEM sample data set. The specific training process will be described below and will not be repeated here.

[0040] Specifically, low-resolution DEM data may be input into the first generator so that the first generator outputs high-resolution DEM data.

[0041] Obviously, since the first generator is obtained by training the second generator based on the high-resolution DEM sample data set, the accuracy of the first generator is improved. In this way, high-resolution DEM data is obtained based on low-resolution DEM data and combined with the first generator, which improves the accuracy of obtaining high-resolution DEM data. In addition, since high-resolution DEM data can be obtained based on the first generator, the difficulty and cost of obtaining high-resolution DEM data are reduced.

[0042] Next, the process of training the second generator according to the high-resolution DEM sample data set to obtain the first generator is described.

[0043] like Figure 2 As shown, the training process may include the following steps:

[0044] 201. Divide the high-resolution DEM sample data set into multiple high-resolution DEM sample data groups.

[0045] The high-resolution DEM sample data set may include multiple high-resolution DEM sample data, and one high-resolution DEM sample data may be high-resolution DEM data of an image. The image may be an image taken based on different terrains, including but not limited to plateau areas, basin areas, plain areas, etc.

[0046] The resolution of the high-resolution DEM sample data may be, for example, 30 meters or 12.5 meters, etc., and this embodiment of the present application does not impose any special limitation on this.

[0047] The high-resolution DEM sample data set may include a plurality of high-resolution DEM sample data of different resolutions to improve the diversity of data in the high-resolution DEM sample data set. Based on this, when the high-resolution DEM sample data set is divided into a plurality of high-resolution DEM sample data groups, the high-resolution DEM sample data of the same resolution may be divided into a group to obtain a plurality of high-resolution DEM sample data groups. Among them, the resolutions of different high-resolution DEM sample data in the same high-resolution DEM sample data group are the same, and the resolutions of the high-resolution DEM sample data in different high-resolution DEM sample data groups are different.

[0048] The number of high-resolution DEM sample data in different high-resolution DEM sample data groups may be the same or different, and the embodiment of the present application does not impose any special limitation on this.

[0049] 202. Downsample multiple high-resolution DEM sample data sets to obtain multiple low-resolution DEM sample data sets.

[0050] A bilinear interpolation method can be used to downsample multiple high-resolution DEM sample data sets to obtain multiple low-resolution DEM sample data sets.

[0051] The multiple high-resolution DEM sample data sets correspond one-to-one to the multiple low-resolution DEM sample data sets.

[0052] The multiple high-resolution DEM sample data in one high-resolution DEM sample data group correspond one-to-one to the multiple low-resolution DEM sample data in the corresponding low-resolution DEM sample data group.

[0053] 203.Build the first training network.

[0054] The first training network includes a second generator, a discriminator, and a first loss function.

[0055] The second generator may include a plurality of deep residual dense modules, a plurality of adaptive residual convolution modules, and a sub-pixel convolution layer.

[0056] Exemplarily, the second generator may include 5 deep residual dense modules, 10 adaptive residual convolution modules and one sub-pixel convolution layer.

[0057] Each deep residual dense module may include multiple dense residual modules. Each dense residual module may include multiple convolutional layers and a first activation function. Among them, multiple convolutional layers are densely connected. The first activation function may use LeakyReLU, for example.

[0058] Exemplarily, each deep residual dense module may include 3 dense residual modules. Each dense residual module may include 5 convolutional layers and a first activation function,

[0059] Each adaptive residual convolution module may include multiple variable convolution layers and a second activation function. The multiple variable convolution layers use jump connections to prevent gradient vanishing and overfitting problems. The second activation function may use ReLU, for example.

[0060] Exemplarily, each adaptive residual convolution module may include 2 variable convolution layers and a second activation function.

[0061] Exemplarily, the adaptive residual convolution module can realize adaptive terrain feature extraction by improving the convolution module of the residual module into a deformable convolution.

[0062] The deep residual dense module can learn the global feature information, so that it can better combine the input DEM data for comprehensive recovery.

[0063] The adaptive residual convolution module can adaptively learn terrain features, thereby expanding the learning scope during the training process.

[0064] The sub-pixel convolution layer can improve the resolution of DEM data and achieve the purpose of upsampling.

[0065] Obviously, since the second generator includes multiple deep residual dense modules, multiple adaptive residual convolution modules and sub-pixel convolution layers, and the deep residual dense modules can learn global feature information, the adaptive residual convolution modules can adaptively learn terrain features, and the sub-pixel convolution layers can improve the resolution of DEM data. Therefore, the second generator can improve the resolution of DEM data. Moreover, since the deep residual dense modules can learn global feature information and the adaptive residual convolution modules can adaptively learn terrain features, during the training process of the second generator, the diversity and accuracy of the second generator learning are improved, thereby improving the accuracy of the first generator obtained by training.

[0066] The discriminator may include a U-Net framework, an attention mechanism gate module and a splicing module, wherein the splicing module may realize information transmission between the U-Net framework and the attention mechanism gate module.

[0067] Since the attention mechanism gate module can expand the receptive field and capture semantic context information, by integrating the attention mechanism gate module in the discriminator, fewer models and a small number of additional model parameters can be used for training. There is no need to crop the region of interest between networks, and the attention mechanism gate module can gradually suppress feature responses in irrelevant background areas.

[0068] 204. Based on the first low-resolution DEM sample data set, train the second generator to obtain a first high-resolution DEM prediction data set.

[0069] The first low-resolution DEM sample data set is one of a plurality of low-resolution DEM sample data sets.

[0070] The first high-resolution DEM prediction data set may include a plurality of high-resolution DEM prediction data.

[0071] The plurality of low-resolution DEM sample data in the first low-resolution DEM sample data group corresponds one-to-one to the plurality of high-resolution DEM prediction data in the first high-resolution DEM prediction data group.

[0072] Specifically, a plurality of low-resolution DEM sample data in the first low-resolution DEM sample data group may be input into the second generator, so that the second generator outputs a plurality of high-resolution DEM prediction data in the first high-resolution DEM prediction data group.

[0073] 205. The first high-resolution DEM prediction data group and the first high-resolution DEM sample data group are discriminated by a discriminator to obtain first discriminant information.

[0074] The first high-resolution DEM sample data group is a high-resolution DEM sample data group corresponding to the first low-resolution DEM sample data group among the multiple high-resolution DEM sample data groups.

[0075] The specific implementation process of 205 can be:

[0076] The high-resolution DEM prediction data corresponding to one low-resolution DEM sample data in the first low-resolution DEM sample data group and the high-resolution DEM sample data corresponding to the first high-resolution DEM prediction data group are input into the discriminator, so that the discriminator outputs a discriminant information corresponding to the low-resolution DEM sample data. In this way, the discriminant information corresponding to each low-resolution DEM sample data in the first low-resolution DEM sample data group is determined to obtain a plurality of discriminant information, and the first discriminant information can be obtained by combining the plurality of discriminant information.

[0077] 206. Determine first evaluation information based on the first discrimination information and the first high-resolution DEM prediction data set, in combination with the first loss function.

[0078] Exemplarily, the first evaluation information may include global evaluation information and adversarial evaluation information. Based on this, the specific implementation process of 206 may be:

[0079] The global evaluation information is determined by the following formula:

[0080] L content =L mse +βL φmse

[0081]

[0082]

[0083] Among them, L content is the global evaluation information, L mse is the first global evaluation information, L φmse is the second global evaluation information, β is 0.006, N1 is the number of high-resolution DEM prediction data in the first high-resolution DEM prediction data group, 1≤i≤N1, I HR is a high-resolution DEM sample data in the first high-resolution DEM sample data group, I LR is a low-resolution DEM sample data in the first low-resolution DEM sample data group, is the second generator, is a high-resolution DEM prediction data in the first high-resolution DEM prediction data group, φ(I HR ) is I HR The characteristic data of for feature data.

[0084] It should be noted that φ(I HR )and It can be obtained in the above process of training the second generator.

[0085] The adversarial evaluation information can be determined by the following formula:

[0086]

[0087] Among them, L Gen is the adversarial evaluation information, N1 is the number of high-resolution DEM prediction data in the first high-resolution DEM prediction data group, 1≤i≤N1, D is the discriminator, It is a high-resolution DEM prediction data in the first high-resolution DEM prediction data group. It is one of the first discrimination information.

[0088] It should be noted that the above formula can be understood as the first loss function.

[0089] 207. Adjust the values ​​of the parameters in the second generator and the discriminator according to the first evaluation information, and iteratively train the second generator, the discriminator and the first loss function after the adjusted parameter values ​​based on the second low-resolution DEM sample data group until the number of iterations is reached.

[0090] The second low-resolution DEM sample data group is one of a plurality of low-resolution DEM sample data groups, and the second low-resolution DEM sample data group is different from the first low-resolution DEM sample data group.

[0091] It should be noted that the training process of 207 can refer to the training process of 204 to 207, which will not be described here. The number of iterations can be determined according to the number of multiple low-resolution DEM sample data groups. For example, the number of iterations can be 100 or 150, etc., and this embodiment of the application does not make any special limitation on this.

[0092] 208. The second generator obtained by the last iterative training is determined as the first generator.

[0093] Obviously, the first training network includes a second generator, a discriminator and a first loss function, and from the above training process, it can be seen that the training of the second generator is completed by cooperating with each other, and by adjusting the values ​​of the parameters in the second generator and the discriminator, thereby improving the accuracy of the training.

[0094] In some embodiments, Figure 3 As shown, before constructing the first training network, the following steps may also be included:

[0095] 301. Construct a second training network; wherein the second training network includes a third generator and a second loss function.

[0096] The structure of the third generator can refer to the structure of the second generator, and will not be repeated here.

[0097] 302. According to the third low-resolution DEM sample data set, the third generator is trained to obtain a third high-resolution DEM prediction data set.

[0098] The third low-resolution DEM sample data group is one of the multiple low-resolution DEM sample data groups, and the third low-resolution DEM sample data group is different from the second low-resolution DEM sample data group and the first low-resolution DEM sample data group.

[0099] It should be noted that the implementation process of 302 can refer to the implementation process of 204, which will not be repeated here.

[0100] 303. Determine second evaluation information based on the third high-resolution DEM prediction data set and the third high-resolution DEM sample data set, and in combination with the second loss function.

[0101] The third high-resolution DEM sample data group is a high-resolution DEM sample data group corresponding to the third low-resolution DEM sample data group among the multiple high-resolution DEM sample data groups.

[0102] Exemplarily, the second evaluation information may be determined by the following formula:

[0103]

[0104] Among them, L Gen is the second evaluation information, N2 is the number of high-resolution DEM prediction data in the third high-resolution DEM prediction data group, 1≤j≤N2, h j is the jth high-resolution DEM prediction data in the third high-resolution DEM prediction data set, It is the jth high-resolution DEM sample data in the third high-resolution DEM sample data group.

[0105] It should be noted that the above formula can be understood as the second loss function.

[0106] 304. Adjust the values ​​of the parameters in the third generator according to the second evaluation information, and iteratively train the third generator with the adjusted parameter values ​​and the second loss function based on the fourth low-resolution DEM sample data group until the iteration number is reached.

[0107] The fourth low-resolution DEM sample data group is one of multiple low-resolution DEM sample data groups, and the fourth low-resolution DEM sample data group is different from the first low-resolution DEM sample data group, the second low-resolution DEM sample data group and the third low-resolution DEM sample data group.

[0108] It should be noted that the training process here can refer to the implementation process of 302 and 303, and will not be repeated here.

[0109] 305. The third generator obtained by the last iterative training is determined as the second generator.

[0110] Obviously, from the above training process, it can be seen that the third generator and the second loss function cooperate with each other and the third generator is trained by adjusting the values ​​of the parameters in the third generator, which improves the accuracy of the training and thus improves the accuracy of the trained second generator. In this way, constructing the first training network based on the second generator trained here is equivalent to using the pre-trained second generator. In this way, when training the second generator in the first training network, the number of training times can be reduced, and the training efficiency and training accuracy can be improved.

[0111] In some embodiments, Figure 4 As shown, the method for obtaining the high-resolution DEM data may also include:

[0112] 401. After completing the last iteration training of the second generator, a high-resolution DEM verification dataset is obtained.

[0113] For the description of the high-resolution DEM validation dataset, please refer to the description of the high-resolution DEM sample dataset above, which will not be repeated here.

[0114] 402. Downsample the high-resolution DEM validation dataset to a low-resolution DEM validation dataset.

[0115] 403. Based on the low-resolution DEM validation data set and in combination with the second generator obtained by the last iteration training, a high-resolution DEM validation prediction set is determined.

[0116] 404. According to the high-resolution DEM verification prediction set and the high-resolution DEM verification data set, determine whether the second generator obtained by the last iteration training passes the verification.

[0117] Exemplarily, when the difference between the high-resolution DEM verification prediction set and the high-resolution DEM verification data set meets the requirements, it is determined that the verification has passed; otherwise, it is determined that the verification has failed.

[0118] 405. If so, the second generator obtained by the last iterative training is determined as the first generator.

[0119] 406. If not, adjust the structure of the first training network.

[0120] Obviously, by downsampling the high-resolution DEM verification data set to a low-resolution DEM verification data set, and determining the high-resolution DEM verification prediction set based on the low-resolution DEM verification data set and the second generator obtained by the last iterative training, the second generator obtained by the last iterative training is verified based on the high-resolution DEM verification prediction set and the high-resolution DEM verification data set. If the verification is passed, the second generator obtained by the last iterative training is determined as the first generator. If the verification is not passed, the structure of the first training network is adjusted to train the readjusted first training network, thereby improving the accuracy of the first training network.

[0121] like Figure 5 As shown, an electronic device 500 provided by an embodiment of the present invention includes a memory 510 and a processor 520; the memory 510 is used to store a computer program; the processor 520 is used to implement the method for obtaining high-resolution DEM data as described above when executing the computer program.

[0122] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for obtaining high-resolution DEM data as described above is implemented.

[0123] An electronic device 500 that can be used as a server or client of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device 500 is intended to represent various forms of digital electronic computer equipment, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 500 can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples, and are not intended to limit the implementation of the present invention described and / or required herein.

[0124] The electronic device 500 includes a computing unit, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The computing unit, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0125] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc. In the present application, the unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment of the present invention. In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0126] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A method for obtaining high-resolution DEM data, characterized in that: include: Obtain low-resolution DEM data; According to the low-resolution DEM data, and in combination with the first generator, high-resolution DEM data is obtained; The first generator is obtained by training the second generator based on a high-resolution DEM sample data set.

2. The method according to claim 1, characterized in that The method further comprises: The second generator is trained according to the high-resolution DEM sample data set to obtain the first generator; wherein: The training of the second generator according to the high-resolution DEM sample data set comprises: Divide the high-resolution DEM sample data set into multiple high-resolution DEM sample data groups; Downsampling the plurality of high-resolution DEM sample data groups to obtain a plurality of low-resolution DEM sample data groups; Wherein, the plurality of high-resolution DEM sample data groups correspond one-to-one to the plurality of low-resolution DEM sample data groups; Constructing a first training network; wherein the first training network includes the second generator, a discriminator and a first loss function; Training the second generator according to the first low-resolution DEM sample data set to obtain a first high-resolution DEM prediction data set; Wherein, the first low-resolution DEM sample data group is one of the multiple low-resolution DEM sample data groups; The first high-resolution DEM prediction data group and the first high-resolution DEM sample data group are discriminated by the discriminator to obtain first discriminant information; The first high-resolution DEM sample data group is a high-resolution DEM sample data group corresponding to the first low-resolution DEM sample data group among the multiple high-resolution DEM sample data groups; Determining first evaluation information based on the first discriminant information and the first high-resolution DEM prediction data group in combination with the first loss function; Adjusting the values ​​of the parameters in the second generator and the discriminator according to the first evaluation information, and iteratively training the second generator, the discriminator and the first loss function after adjusting the values ​​of the parameters based on a second low-resolution DEM sample data group until the number of iterations is reached; The second low-resolution DEM sample data group is one of the multiple low-resolution DEM sample data groups, and the second low-resolution DEM sample data group is different from the first low-resolution DEM sample data group; The second generator obtained by the last iterative training is determined as the first generator.

3. The method according to claim 2, characterized in that Before constructing the first training network, the method further includes: Constructing a second training network; wherein the second training network includes a third generator and a second loss function; According to the third low-resolution DEM sample data set, the third generator is trained to obtain a third high-resolution DEM prediction data set; The third low-resolution DEM sample data group is one of the multiple low-resolution DEM sample data groups, and the third low-resolution DEM sample data group is different from the second low-resolution DEM sample data group and the first low-resolution DEM sample data group; Determining second evaluation information based on the third high-resolution DEM prediction data group and the third high-resolution DEM sample data group and in combination with the second loss function; The third high-resolution DEM sample data group is a high-resolution DEM sample data group corresponding to the third low-resolution DEM sample data group among the multiple high-resolution DEM sample data groups; Adjusting the values ​​of the parameters in the third generator according to the second evaluation information, and iteratively training the third generator and the second loss function after adjusting the values ​​of the parameters based on a fourth low-resolution DEM sample data group until the number of iterations is reached; The fourth low-resolution DEM sample data group is one of the multiple low-resolution DEM sample data groups, and the fourth low-resolution DEM sample data group is different from the first low-resolution DEM sample data group, the second low-resolution DEM sample data group and the third low-resolution DEM sample data group; The third generator obtained by the last iterative training is determined as the second generator.

4. The method according to claim 2, characterized in that: The method further comprises: After completing the last iteration training of the second generator, a high-resolution DEM verification dataset is obtained; Downsampling the high-resolution DEM verification dataset into a low-resolution DEM verification dataset; Determine a high-resolution DEM verification prediction set based on the low-resolution DEM verification data set and the second generator obtained by the last iteration training; According to the high-resolution DEM verification prediction set and the high-resolution DEM verification data set, determining whether the second generator obtained by the last iteration training passes the verification; If yes, the second generator obtained by the last iteration training is determined as the first generator; If not, the structure of the first training network is adjusted.

5. The method according to any one of claims 1 to 4, characterized in that: The second generator includes multiple deep residual dense modules, multiple adaptive residual convolution modules and sub-pixel convolution layers.

6. The method according to claim 5, characterized in that Each of the deep residual dense modules includes a plurality of dense residual modules, each of the dense residual modules includes a plurality of convolutional layers and a first activation function; Wherein, the multiple convolutional layers are densely connected; Each of the adaptive residual convolution modules includes a plurality of variable convolution layers and a second activation function; Wherein, the multiple variable convolutional layers adopt jump connections.

7. The method according to claim 2, characterized in that: The discriminator includes a U-Net framework, an attention mechanism gate module and a splicing module; The splicing module is used to realize information transmission between the U-Net framework and the attention mechanism gate module.

8. The method according to claim 2, characterized in that: The first evaluation information includes global evaluation information and adversarial evaluation information; The determining of the first evaluation information according to the first discrimination information and the first high-resolution DEM prediction data group in combination with the first loss function includes: The global evaluation information is determined by the following formula: L content =L mse +βL φmse Among them, L content is the global evaluation information, L mse is the first global evaluation information, L φmse is the second global evaluation information, β is 0.006, N1 is the number of high-resolution DEM prediction data in the first high-resolution DEM prediction data group, 1≤i≤N1, I HR is a high-resolution DEM sample data in the first high-resolution DEM sample data group, I LR is a low-resolution DEM sample data in the first low-resolution DEM sample data group, For the second generator, is a high-resolution DEM prediction data in the first high-resolution DEM prediction data group, φ(I HR ) is I HR The characteristic data of for characteristic data; The adversarial evaluation information is determined by the following formula: Among them, L Gen is the adversarial evaluation information, N1 is the number of high-resolution DEM prediction data in the first high-resolution DEM prediction data group, 1≤i≤N1, D is the discriminator, It is a high-resolution DEM prediction data in the first high-resolution DEM prediction data group.

9. The method according to claim 3, characterized in that: The determining of the second evaluation information based on the third high-resolution DEM prediction data set and the third high-resolution DEM sample data set in combination with the second loss function includes: The second evaluation information is determined by the following formula: Among them, L Gen is the second evaluation information, N2 is the number of high-resolution DEM prediction data in the third high-resolution DEM prediction data group, 1≤j≤N2, h j is the j-th high-resolution DEM prediction data in the third high-resolution DEM prediction data group, It is the j-th high-resolution DEM sample data in the third high-resolution DEM sample data group.

10. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is configured to implement the method for obtaining high-resolution DEM data according to any one of claims 1 to 9 when executing the computer program.

Images