DEM (Digital Elevation Model) super-resolution reconstruction method and system fusing gradient feature Transform

Abstract

The invention provides a DEM super-resolution reconstruction method and system fusing gradient feature Transform, and belongs to the technical field of image processing, the reconstruction method is composed of a plurality of residual feature fusion blocks, a multi-scale pyramid segmentation Transform layer in the residual feature fusion blocks fuses multi-head self-attention and pyramid segmentation attention, and the multi-scale pyramid segmentation Transform layer fuses the multi-head self-attention and the pyramid segmentation attention. In order to make the reconstructed DEM closer to reality, a double-filtering convolution block is used, high and low frequency features of the DEM are fully extracted through a parallel structure of two different filtering blocks, and finally super-resolution reconstruction is carried out under double constraints of a gradient domain and a height domain. The method adapts to topographic features of different scales and complexities, and solves the problem that DEM details cannot be accurately recovered by a traditional algorithm.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing and relates to a DEM super-resolution reconstruction method and system integrating a gradient feature Transformer. Background Art

[0002] DEM is a commonly used terrain data model in geographic information system (GIS). DEM plays an important role in various geographic analyses and is widely used in terrain analysis, hydrological modeling, terrain-assisted navigation, and environmental simulation. The higher the DEM accuracy, the more accurate the results of geographic analysis. Therefore, how to obtain high-resolution DEM has become a research hotspot. Methods such as vehicle-mounted lidar measurement, aerial photogrammetry, and satellite image measurement are all methods for directly obtaining DEM, but they are time-consuming and labor-intensive, and are not suitable for large-scale collection methods in some places. And with the increase in DEM measurement methods, there are more and more DEMs for the same area, showing multi-resolution and multi-phase characteristics. Researchers have begun to explore the use of multi-source DEM for data fusion to improve DEM resolution. However, due to the confidentiality of DEM in some areas, there are certain difficulties in obtaining multi-source DEM. Therefore, the current mainstream method is to super-resolution reconstruct low-resolution DEM to obtain high-resolution DEM.

[0003] However, the use of set projection method, iterative back-projection method, maximum a posteriori estimation method and various interpolation methods cannot accurately obtain DEM detail information, resulting in the loss of terrain edges, details and clarity in the reconstructed DEM. With the continuous development of deep learning, the super-resolution model based on convolutional neural network (CNN) can abstract the mapping relationship between high-resolution data and low-resolution data learned, and realize super-resolution reconstruction. Therefore, researchers have focused on applying super-resolution reconstruction methods based on deep learning to DEM. In 2016, Chen et al., based on CNN, output high-resolution DEM through an optimization method that minimizes error. The DEM reconstructed by this method achieved better results than traditional interpolation methods. In 2018, Xu et al. used deep convolutional networks combined with transfer learning to successfully transfer the knowledge of natural image learning to the field of DEM super-resolution reconstruction, achieving high-precision results with a small number of DEMs. In 2020, Jiao et al. proposed a DEM super-resolution reconstruction model based on a deep residual network, which optimizes the model by minimizing the reconstruction error, thereby improving the resolution and accuracy of the DEM. In 2022, Lin et al. introduced a DEM super-resolution reconstruction model that combines internal and external learning. The model uses the internal prior knowledge of DEM data and the learning of external image datasets to improve the quality and details of the reconstructed DEM. In 2023, Zhou et al. combined terrain features with CNN and introduced a super-resolution deep model MTF-SR based on multiple terrain features. From the perspective of vector and raster terrain features, it improves the overall accuracy of DEM and retains the advantages of terrain features.

[0004] The above methods are all based on CNN to take DEM as an extension of ordinary images and use grayscale values ​​to represent the elevation information of DEM. However, the spatial information of DEM is extremely complex, and there is a long-term dependency between terrain features, which CNN cannot fully utilize, resulting in the inability to reliably reconstruct the geographic structure details of DEM. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a DEM super-resolution reconstruction method and system integrating gradient feature Transformer, which enhances the ability to obtain global information and model long-distance dependency, fully extracts the high- and low-frequency features of DEM, and realizes super-resolution reconstruction under the dual constraints of gradient domain and height domain.

[0006] The present invention is achieved through the following technical solutions: A DEM super-resolution reconstruction method integrating gradient feature Transformer, including: The high-resolution digital elevation model D is processed to obtain the gradient map of the low-resolution digital elevation model ; Gradient map for low-resolution digital elevation models Perform shallow feature extraction to obtain feature maps ; For feature maps After deep feature extraction through multiple residual feature fusion blocks and dual filter convolution blocks, the feature map is obtained. ; The feature map After being input into the Upscale module, the high-resolution digital elevation model gradient map is obtained through the convolution layer. , for high-resolution digital elevation model gradient maps Under the dual constraints of gradient domain and height domain, a high-resolution digital elevation model is finally obtained. , realizing super-resolution reconstruction of DEM.

[0007] Preferably, a high-resolution digital elevation model The processing process is: S1: High-resolution digital elevation model Divide into several submodules, 90% of which are used as training sets , 5% as validation set , 5% as test set ; S2: High-resolution digital elevation model The submodule used for training set in the training set obtains the low-resolution digital elevation model by downsampling method. ; S3: Convert low-resolution digital elevation models The image gradient operator is used to obtain the gradient map of the low-resolution digital elevation model. .

[0008] Preferably, the downsampling method is: High-resolution digital elevation models The submodules used for the training set are downsampled by factors of ×2, ×3, and ×4 to obtain the corresponding low-resolution digital elevation models. .

[0009] Preferably, the gradient map of the low-resolution digital elevation model use Convolution performs shallow feature extraction.

[0010] Preferably, the process of deep feature extraction is: S1, the obtained The features are deeply extracted, where the deep feature extraction module contains 5 residual feature fusion blocks and 1 dual filter convolution block; S2, feature map Enter the first residual feature fusion block, extract the high and low frequency features in the feature map, and then use dense connection to perform symmetrical residual links between modules to achieve mutual fusion of features at different stages and output the feature map. ; S3, the feature map Input it to the next residual feature fusion block, repeat the learning steps in S2, and record the obtained feature map as , and then input it into the next residual feature fusion block, and repeat the five RFFBs step by step to obtain the feature map ; S4, Input into the double filter convolution block, we get And compared with the feature map extracted by shallow features Perform residual connection to obtain feature map .

[0011] Preferably, each residual feature fusion block is composed of a multi-scale pyramid segmentation Transformer layer and an STL, which are used alternately; Each residual feature fusion block consists of 3 STL modules and 2 multi-scale pyramid segmentation Transformer layer modules; The multi-scale pyramid segmentation Transformer layer is composed of a fusion of multi-head self-attention and pyramid segmentation attention.

[0012] Preferably, the dual-filter convolution block adopts a parallel structure of two different filter blocks, including an upper filter block and a lower filter; by fusing the feature maps of the two filter blocks, the integration of DEM information of different scales is achieved; The upper filter block uses a 3x3 convolution kernel, combined with a Relu activation function and a residual link, while the lower filter block uses a 7x7 convolution kernel.

[0013] A DEM super-resolution reconstruction system integrating gradient feature Transformer, including: Image processing module for converting high-resolution digital elevation models After processing, the gradient map of the low-resolution digital elevation model is obtained ; Shallow feature extraction module for gradient maps of low-resolution digital elevation models Perform shallow feature extraction to obtain feature maps , Deep feature extraction module, used to extract feature maps After deep feature extraction through multiple residual feature fusion blocks and dual filter convolution blocks, the feature map is obtained. ; Super-resolution reconstruction module is used to transform feature maps After being input into the Upscale module, the high-resolution digital elevation model gradient map is obtained through the convolution layer. ; Height domain reconstruction module, high resolution digital elevation model gradient map Obtaining high-resolution digital elevation model under dual constraints in gradient domain and height domain ..

[0014] An electronic device comprises a processor and a memory, wherein at least one instruction is stored in the memory, and the instruction stored in the memory is executed to implement the DEM super-resolution reconstruction method of the gradient feature fused Transformer as described.

[0015] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the DEM super-resolution reconstruction method of the gradient feature fused Transformer is implemented.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention provides a DEM super-resolution reconstruction method and system integrating gradient feature Transformer. In order to enhance the global information acquisition capability and long-distance dependency modeling capability, the model is composed of a plurality of residual feature fusion blocks. The multi-scale pyramid segmentation Transformer layer inside the model integrates the multi-head self-attention and the pyramid segmentation attention, so as to better capture the local and overall feature information of the DEM, and solve the problem of detail loss and distortion caused by the inability to fully utilize the complex terrain features of the DEM. Secondly, in order to make the reconstructed DEM closer to reality, a dual-filter convolution block is used. Through the parallel structure of two different filter blocks, the high and low frequency features of the DEM are fully extracted. Finally, super-resolution reconstruction is performed under the dual constraints of the gradient domain and the height domain, so that the reconstructed DEM restores the real landform. The present invention solves the problem of geometric clue differences between elevation features and RGB features caused by expanding the DEM as a two-dimensional image.

[0017] Furthermore, the present invention uses PSNR and ARE as objective evaluation indicators and introduces slope and aspect as additional evaluation criteria for the reconstruction effect of terrain elements. By combining MSA and PSA, multi-scale feature extraction and global context modeling are achieved and complex terrain features are effectively captured. After further integration through a multi-layer perceptron (MLP), local information can be extracted, so that the model can better understand the detailed terrain features in the DEM data.

[0018] Furthermore, the present invention intuitively introduces the parallel structure of two different filter blocks of DFCB to consider high-frequency and low-frequency features at the same time, which can capture the multi-scale information of the terrain model more comprehensively. One branch uses a 3x3 convolution kernel, combined with activation functions and residual links, focusing on accurately extracting high-frequency information of DEM and capturing small-scale features such as rivers, valleys and mountains. The other branch uses a 7x7 convolution kernel, combined with activation functions and residual links, aiming to obtain a wider range of low-frequency information, covering large-scale features such as mountains and plateaus. By fusing the feature maps of these two branches, DFCB realizes the integration of DEM information of different scales, improving the model's comprehensive perception and accurate expression of DEM terrain features.

[0019] Furthermore, the digital elevation model reconstructed by super-resolution in the present invention is more accurate, closest to the original image, and has better effects than the comparison algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the structure of the method of the present invention; Figure 2 It is a schematic diagram of the MPSA structure of the present invention; Figure 3 It is a schematic diagram of the experimental area of ​​the present invention; Figure 4 It is a schematic diagram of the fusion of high and low frequency features of DFCB of the present invention; Figure 5 This is a comparison chart of the DEM super-resolution reconstruction effect of the present invention and other methods; Figure 6 This is a comparison chart of the slope super-resolution reconstruction effects of the present invention and other methods. DETAILED DESCRIPTION

[0021] The present invention is further described in detail below in conjunction with specific embodiments, which are intended to explain the present invention rather than to limit it.

[0022] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0023] Example 1 A DEM super-resolution reconstruction method integrating gradient feature Transformer, comprising: Gradient extraction steps: S1: High-resolution digital elevation model Divide into several blocks of 256*256 size, 90% of which are used as training sets , 5% as validation set , 5% as test set ; S2: High-resolution digital elevation model for training set , a low-resolution digital elevation model is obtained by downsampling ; The high-resolution digital elevation model of the training set , downsampled by factors of ×2, ×3, and ×4 to obtain the corresponding low-resolution digital elevation model ; S3: Low-resolution digital elevation model Use the image gradient operator to obtain the gradient map of the low-resolution digital elevation model ; Shallow feature extraction steps: S1: Gradient map for low-resolution digital elevation model use Convolution performs shallow feature extraction to obtain features ; Deep feature extraction steps: S1: The obtained The deep feature extraction module contains five residual feature fusion blocks (RFFB) and one dual filtering convolutional block (DFCB). S2: Enter the first RFFB, extract the high and low frequency features in the feature map, and then use dense connection to perform symmetrical residual links between modules to achieve mutual fusion of features at different stages and output the feature map ; S3: Feature map Input it into the next RFFB, repeat the learning steps in S2, and record the obtained feature map as , and then input it into the next RFFB, and repeat the step-by-step execution of 5 RFFBs to obtain the feature map ; S4: Enter into DFCB and get And compared with the features extracted by shallow features Perform residual connection to obtain feature map ; In the deep feature extraction step, each RFFB is composed of a Multi-Pyramid Segmented Transformer Layer (MPSTL) and a Swin Transformer Layer (STL), which are used alternately to balance the needs of processing local sensitivity and global dependency of DEM data; In the deep feature extraction step, MPSTL is composed of a fusion of Multi-head Self-attention (MSA) and Pyramid Split Attention (PSA); by combining Multi-head Self-attention (MSA) and Pyramid Split Attention (PSA), multi-scale feature extraction and global context modeling are achieved and complex terrain features are effectively captured; The specific fusion process is as follows: the input of MPSTL is , the output process is as follows:

[0024]

[0025] in is the layer normalization function, is the pyramid segmentation attention function, is the multi-head self-attention function; The calculation is as follows: Input Window ,query, key, value matrix The calculation formula is: ; in , and , is the learning weight.

[0026] Self-attention process for: ; The PSA module divides the input feature map into S groups of convolutions. The convolution kernel of each group gradually increases. When the convolution kernel is large, group convolution is performed. The feature map of each group is calculated as:

[0027] After convolution with different convolution kernels, each group of features is concatenated to obtain a multi-scale preprocessed feature map:

[0028] Next, the SEWeight module is used to obtain the attention weights of feature maps of different scales. The normalized channel weights are:

[0029] Finally, the output feature map is obtained by performing XOR operation on the feature map and the corresponding channel weight. :

[0030] in Represents feature connection.

[0031] In the deep feature extraction step, each RFFB consists of 5 layers, where It is RFFB STL, It is RFFB MPSTL; DFCB exists in each RFFB and after the last RFFB; each RFFB has 3 STL and 2 MPSTL modules; DFCB adopts a parallel structure of two different filter blocks. The upper filter block uses a 3x3 convolution kernel, combined with Relu activation function and residual link, and the lower filter block uses a 7x7 convolution kernel. By fusing the feature maps of these two filter blocks, DFCB realizes the integration of DEM information at different scales. Super-resolution reconstruction steps: S1: Feature map After being input into the Upscale module, the high-resolution digital elevation model gradient map is obtained through the convolution layer. ; Height domain reconstruction steps: S1: High-resolution digital elevation model gradient map Under the dual constraints of gradient domain and height domain, a high-resolution digital elevation model is finally obtained. .

[0032] High-resolution digital elevation model gradient map Under the dual constraints of gradient domain and height domain, a high-resolution digital elevation model is finally obtained. .

[0033] A DEM super-resolution reconstruction system integrating gradient feature Transformer, including: Image processing module for converting high-resolution digital elevation models After processing, the gradient map of the low-resolution digital elevation model is obtained ; Shallow feature extraction module for gradient maps of low-resolution digital elevation models Perform shallow feature extraction to obtain feature maps , Deep feature extraction module, used to extract feature maps After deep feature extraction through multiple residual feature fusion blocks and dual filter convolution blocks, the feature map is obtained. ; Super-resolution reconstruction module is used to transform feature maps After being input into the Upscale module, the high-resolution digital elevation model gradient map is obtained through the convolution layer. , Height domain reconstruction module, high resolution digital elevation model gradient map Obtaining high-resolution digital elevation model under dual constraints in gradient domain and height domain .

[0034] An electronic device comprises a processor and a memory, wherein at least one instruction is stored in the memory, and the instruction stored in the memory is executed to implement the DEM super-resolution reconstruction method of the gradient feature fused Transformer.

[0035] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, a DEM super-resolution reconstruction method such as a gradient feature transformer is implemented.

[0036] A terminal device is provided in one embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0037] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to accomplish the present invention.

[0038] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0039] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0040] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0041] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0042] It should be noted that the terms in the specification and claims of the present invention and the above-mentioned drawings that include a series of steps or units in the process, method, system, product or equipment are not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0043] Example 2 The embodiment of the present invention discloses a DEM super-resolution reconstruction method integrating gradient feature Transformer, such as Figure 1 As shown, it consists of 5 parts: gradient extraction module, shallow feature extraction, deep feature extraction, super-resolution reconstruction and height domain reconstruction.

[0044] (1) Gradient extraction module Convert low-resolution DEM Obtain low-resolution DEM gradient map through edge detection Prewitt operator , and as input, the process is as follows:

[0045] in and They are the horizontal gradient map and the vertical gradient map respectively.

[0046] (2) Shallow feature extraction This module uses Convolution is used to extract shallow features, as shown in formula (2):

[0047] in is the shallow feature of the gradient map, is the number of feature map channels. The use of convolution can enable the network to have shallow feature extraction capabilities and can extract features of low-frequency information in the gradient map.

[0048] (3) Deep feature extraction Will get Features are extracted in depth. The deep feature extraction module contains RFFB and a double filter convolution block. The final output feature map for: As Input features, after RFFB and DFCB are extracted layer by layer, and the output of this process can be expressed as:

[0049]

[0050] (4) Super-resolution reconstruction The super-resolution reconstruction module consists of two convolution blocks, a ReLU function and a sub-pixel convolution layer. When reconstructing the high-resolution gradient map, it is then input into the super-resolution reconstruction module for reconstruction, as shown below:

[0051] in Is the rebuild module function.

[0052] (5) Height domain reconstruction Using the reconstructed high-resolution gradient map , reconstructing a high-resolution DEM in the height domain , as shown below:

[0053] in, represents upsampling, and represents the number of iterations, Represents the divergence operator, which is used to iteratively update the weight parameters of each network layer and use the gradient descent algorithm to obtain the optimal .

[0054] The present invention uses PSNR and ARE as objective evaluation indicators and introduces slope and aspect as additional evaluation criteria for the reconstruction effect of terrain elements. Figure 2 It is a schematic diagram of the MPSA structure. By combining MSA and PSA, multi-scale feature extraction and global context modeling can be achieved and complex terrain features can be effectively captured. After further integration through the Multi-Layer Perceptron (MLP), local information can be extracted, enabling the model to better understand the detailed terrain features in the DEM data.

[0055] Figure 4 The parallel structure of two different filter blocks of DFCB is intuitively introduced to consider both high-frequency and low-frequency features at the same time, which can more comprehensively capture the multi-scale information of the terrain model. One branch uses a 3x3 convolution kernel, combined with activation functions and residual links, focusing on accurately extracting high-frequency information of DEM and capturing small-scale features such as rivers, valleys and mountains. The other branch uses a 7x7 convolution kernel, combined with activation functions and residual links, aiming to obtain a wider range of low-frequency information, covering large-scale features such as mountains and plateaus. By fusing the feature maps of these two branches, DFCB realizes the integration of DEM information at different scales, improving the model's comprehensive perception and accurate expression of DEM terrain features.

[0056] Figure 5 and Figure 6 They are the reconstruction effect diagrams of the digital elevation models with different precisions under the zoom factor of 3, Figure 5 and Figure 6 It can be seen that the digital elevation model reconstructed by super-resolution in the present invention is more accurate, closest to the original image, and has better effects than the comparison algorithm.

[0057] Comparison algorithm source: [EDSR]: Lim B , Son S , Kim H ,et al.Enhanced Deep Residual Networks for Single Image Super-Resolution[J]. arXiv:1707.02921,2017. [ESRGAN]: Wang [RCAN]: Zhang Y , Li K , Li K ,et al.Image Super-Resolution Using Very DeepResidual Channel Attention Networks[J].arXiv:1807.02758,2018. [SwinIR]: Jingyun Liang, Jiezhang Cao, Guolei Sun, et al. SwinIR: Image Restoration Using Swin Transformer [J].arXiv:2108.10257. The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any ordinary technician in the industry can smoothly implement the present invention as shown in the drawings and described above. However, any equivalent changes, modifications and evolutions made by technicians familiar with the profession without departing from the scope of the technical solution of the present invention using the technical content disclosed above are all equivalent embodiments of the present invention. At the same time, any equivalent changes, modifications and evolutions made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the technical solution of the present invention.

Claims

1. A DEM super-resolution reconstruction method integrating gradient feature Transformer, characterized in that: include, High-resolution digital elevation models After processing, the gradient map of the low-resolution digital elevation model is obtained ; Gradient map for low-resolution digital elevation models Perform shallow feature extraction to obtain feature maps ; For feature maps After deep feature extraction through multiple residual feature fusion blocks and dual filter convolution blocks, the feature map is obtained. ; The residual feature fusion block contains a multi-scale pyramid segmentation Transformer layer, and the multi-scale pyramid segmentation Transformer layer is formed by fusing multi-head self-attention and pyramid segmentation attention; The feature map After being input into the Upscale module, the high-resolution digital elevation model gradient map is obtained through the convolution layer. , for high-resolution digital elevation model gradient maps Under the dual constraints of gradient domain and height domain, a high-resolution digital elevation model is finally obtained. , realizing super-resolution reconstruction of DEM.

2. According to the DEM super-resolution reconstruction method of the gradient feature Transformer fusion method in claim 1, it is characterized in that: High-resolution digital elevation model The processing process is: S1: High-resolution digital elevation model Divide into several submodules, 90% of which are used as training sets , 5% as validation set , 5% as test set ; S2: High-resolution digital elevation model The submodule used for training set in the training set obtains the low-resolution digital elevation model by downsampling method. ; S3: Convert low-resolution digital elevation models The image gradient operator is used to obtain the gradient map of the low-resolution digital elevation model. .

3. According to the DEM super-resolution reconstruction method of the gradient feature transformer fusion according to claim 2, it is characterized in that: The downsampling method is: High-resolution digital elevation models The submodules used for the training set are downsampled by factors of ×2, ×3, and ×4 to obtain the corresponding low-resolution digital elevation models. .

4. The DEM super-resolution reconstruction method of the gradient feature transformer fusion according to claim 1 is characterized in that: The gradient map of the low-resolution digital elevation model use Convolution performs shallow feature extraction.

5. The DEM super-resolution reconstruction method of the gradient feature transformer fusion according to claim 1 is characterized in that: The process of deep feature extraction is: S1, the feature map Perform deep extraction, where the deep feature extraction module contains 5 residual feature fusion blocks and 1 dual filter convolution block; S2, feature map Enter the first residual feature fusion block, extract the high and low frequency features in the feature map, and then use dense connection to perform symmetrical residual links between modules to achieve mutual fusion of features at different stages and output the feature map. ; S3, the feature map Input it to the next residual feature fusion block, repeat the learning steps in S2, and record the obtained feature map as , and then input it into the next residual feature fusion block, and repeat the five RFFBs step by step to obtain the feature map ; S4, Input into the double filter convolution block, we get And compared with the feature map extracted by shallow features Perform residual connection to obtain feature map .

6. The DEM super-resolution reconstruction method of claim 1, characterized in that: Each residual feature fusion block consists of a multi-scale pyramid segmentation Transformer layer and STL, which are used alternately; Each residual feature fusion block consists of 3 STL modules and 2 multi-scale pyramid segmentation Transformer layer modules.

7. The DEM super-resolution reconstruction method of integrating gradient feature Transformer according to claim 1 is characterized in that: The dual-filter convolution block adopts a parallel structure of two different filter blocks, including an upper filter block and a lower filter block; by fusing the feature maps of the two filter blocks, the integration of DEM information of different scales is achieved; The upper filter block uses a 3x3 convolution kernel, combined with a Relu activation function and a residual link, while the lower filter block uses a 7x7 convolution kernel.

8. A DEM super-resolution reconstruction system integrating gradient feature transformer, based on the reconstruction method according to any one of claims 1 to 7, comprising: Image processing module for converting high-resolution digital elevation models After processing, the gradient map of the low-resolution digital elevation model is obtained ; Shallow feature extraction module for gradient maps of low-resolution digital elevation models Perform shallow feature extraction to obtain feature maps , Deep feature extraction module, used to extract feature maps After deep feature extraction through multiple residual feature fusion blocks and dual filter convolution blocks, the feature map is obtained. ; Super-resolution reconstruction module is used to transform feature maps After being input into the Upscale module, the high-resolution digital elevation model gradient map is obtained through the convolution layer. ; Height domain reconstruction module, high resolution digital elevation model gradient map Obtaining high-resolution digital elevation model under dual constraints in gradient domain and height domain .

9. An electronic device, characterized in that: The invention comprises a processor and a memory, wherein at least one instruction is stored in the memory, and the instruction stored in the memory is executed to implement the DEM super-resolution reconstruction method of the gradient feature fused Transformer as described in claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the DEM super-resolution reconstruction method of the gradient feature Transformer fusion as described in claims 1-7 is implemented.

Images