A method for unifying the resolution of variable-resolution images

Abstract

The present invention belongs to the field of image processing, and particularly relates to a method for unifying the resolution of variable-resolution images, aiming to solve the problems that it is difficult to simultaneously maintain the visual quality and information integrity of images, and it is easy to lose detailed information and reduce the utilization rate of images. The present invention includes: cutting out a plurality of segmented images along the edge part of the initial image, and ensuring that there is an overlapping area between adjacent segmented images; inputting them into a super-resolution reconstruction model to obtain high-resolution segmented images, performing geometric correction processing, correcting all segmented images into a unified coordinate system, performing weighted fusion processing on the overlapping areas of the geometrically corrected segmented images, and achieving seamless stitching with consistent tone based on the weight distribution principle of the distance from the center of the segmented image to generate a unified-resolution image. The present invention avoids the loss of detailed information caused by global downsampling / upsampling in traditional resampling methods and overcomes the problem of image blurring caused by a single resampling operation.

Description

Background Art

[0002] Variable-resolution images refer to images with an extremely wide imaging width, capable of capturing a large geographical area at once. The image coverage range far exceeds that of conventional satellites and can cover thousands of kilometers or more. Variable-resolution images greatly improve the efficiency of data acquisition and are crucial for applications such as environmental monitoring, urban planning, and disaster response. Due to factors such as imaging methods, the curvature of the Earth, and changes in the exterior orientation elements of sensors, large-width remote sensing images often exhibit deformation phenomena. This deformation results in a higher resolution at the center (sub-satellite point) of the image and a lower resolution at the edge part.

[0003] If a resampling method is simply adopted, downsampling the high-resolution area and upsampling the low-resolution area to achieve resolution uniformity, this method only considers the requirement of uniform resolution while ignoring the visual perception of the image. At the same time, it may also lead to the loss of detailed information, thereby reducing the utilization rate of the image. This resampling method is only applicable to ordinary satellite remote sensing images with relatively small differences in the resolution of a single image.

[0004] Therefore, for the resolution uniformity of variable-resolution images, more refined image processing techniques need to be adopted to maintain the visual quality and information integrity of the images.

[0005] Based on this, the present invention proposes a method for unifying the resolution of variable-resolution images. Summary of the Invention

[0006] To solve the above problems in the prior art, that is, it is difficult to simultaneously maintain the visual quality and information integrity of the image, which easily leads to the loss of detailed information and the reduction of the image utilization rate, the present invention provides a method for unifying the resolution of variable-resolution images. The method includes the following steps:

[0007] Step S1, obtain the remote sensing image with variable resolution to be unified as the initial image. Taking the center of the initial image as the starting point, based on the change amplitude of the resolution from the center to the edge, cut out multiple piecewise images along the edge part of the initial image, and ensure that there is an overlapping area between adjacent piecewise images;

[0008] Step S2, use the piecewise image corresponding to the low-resolution area as the input data and input it into the pre-constructed and trained super-resolution reconstruction model to obtain a high-resolution piecewise image with the same resolution as the piecewise image at the center;

[0009] Step S3, perform geometric correction processing on the piecewise images after resolution unification, and correct all piecewise images into a unified coordinate system;

[0010] Step S4, perform weighted fusion processing on the overlapping areas of the geometrically corrected segmented images, and achieve seamless stitching with consistent tones based on the weight assignment principle of the distances from the centers of the segmented images, and generate a unified resolution image.

[0011] Further, cut out multiple segmented images along the edge part of the initial image. The method is: starting from the center of the initial image, advance the cutting towards the edge of the initial image, and keep the cutting path consistent with the gradient direction of the resolution change.

[0012] Further, the size of the segmented image is dynamically adjusted according to the resolution change amplitude at its location; among them, the size of the segmented image in the central area is larger than that in the edge area.

[0013] Further, the super-resolution reconstruction model includes a shallow feature extraction module, a deep feature extraction module, and a reconstruction module connected in sequence;

[0014] The shallow feature extraction module extracts primary features through single-layer convolution operations;

[0015] The deep feature extraction module is composed of multiple cascaded residual dense blocks. Each residual dense block contains multiple cascaded residual dense attention modules. The residual dense attention module includes multiple sequentially connected convolutional layers and one spatial attention module, and adjacent residual dense attention modules are connected through layer normalization operations;

[0016] The output of the spatial attention module is added to the input of the residual dense attention module through a residual connection;

[0017] The input of the deep feature extraction module is the primary features output by the shallow feature extraction module, and the output of the deep feature extraction module is deep features;

[0018] The reconstruction module fuses the primary features and the deep features, and after fusion, performs upsampling processing on the fused features through the nearest neighbor interpolation method to generate the final high-resolution image.

[0019] Further, the execution process of the spatial attention module includes:

[0020] Perform depth convolution and pointwise convolution on the input feature map in sequence to generate a spatial attention weight map; multiply the spatial attention weight map with the input feature map channel by channel, and output the weighted feature map.

[0021] Further, in step S3, perform geometric correction processing on the segmented images after resolution unification, and correct all segmented images into a unified coordinate system. The method is:

[0022] Step S31: Construct a rigorous imaging geometric model for satellite remote sensing images;

[0023] Step S32: Based on each piece of image, determine whether the elevation value of the target point can be obtained. If yes, jump to Step S33; otherwise, jump to Step S34.

[0024] Step S33: Based on the rigorous imaging geometric model, given the coordinates of the piece of image, camera parameters, satellite orbit parameters, and the elevation value of the target point, solve the model equation to obtain the object plane coordinates of each target point in the piece of image as the corrected coordinates.

[0025] Step S34: Use the local average elevation of the area covered by the piece of image as the initial value, calculate the initial longitude and latitude, interpolate the elevation value corresponding to the initial longitude and latitude in the DEM data, and substitute it into the imaging geometric positioning model for iterative calculation to obtain the longitude, latitude, and elevation values of the piece of image, and then jump to Step S33.

[0026] Furthermore, substitute it into the imaging geometric positioning model for iterative calculation until the results of two adjacent iterations are less than the preset threshold, and then obtain the longitude, latitude, and elevation values of the piece of image.

[0027] Furthermore, perform weighted fusion processing on the overlapping areas of the geometrically corrected pieces of image. The method is as follows:

[0028] Step S41: For the pixel of any point P in the overlapping area, dynamically assign weights according to its distance from the centers of adjacent pieces of image.

[0029] Step S42: Calculate the pixel of point P based on the weight assignment.

[0030] ;

[0031] Among them, the two pieces of image corresponding to the overlapping area where point P is located are respectively defined as image A and image B. is the pixel value of image A at the position of point P. is the pixel value of image B at the position of point P. is the weight of image A for the pixel of point P. is the weight of image B for the pixel of point P.

[0032] Furthermore, the weight , and its calculation method is as follows:

[0033] ;

[0034] Among them, , are the distances from point P to the right edge and lower edge of image A, , are the distances from point P to the left and upper edges of image B.

[0035] Furthermore, the weight , and its calculation method is as follows:

[0036] ;

[0037] where , are the distances from point P to the right and lower edges of image A, , are the distances from point P to the left and upper edges of image B.

[0038] Advantages of the present invention:

[0039] Resolution adaptive cropping mechanism: By means of a radial cropping method starting from the image center and dynamically adjusting the tile size in combination with the resolution change range, it avoids the loss of detailed information caused by global downsampling / upsampling in the traditional resampling method, retains the resolution distribution characteristics of the original image at the cropping stage, and provides a data basis for subsequent processing.

[0040] Region-based super-resolution reconstruction: Directionally perform super-resolution processing on low-resolution edge tiles, and use a deep learning model to restore high-frequency texture details, effectively maintaining the visual quality of the image while increasing the resolution, and overcoming the problem of image blurring caused by a single resampling operation.

[0041] Geometric consistency correction: Through a geometric correction algorithm based on the extraction of extreme values in the map coordinate system, it eliminates the image deformation caused by the earth's curvature and sensor distortion, ensures the spatial alignment accuracy of all tile images in a unified coordinate system, and provides a geometric basis for seamless stitching.

[0042] Intelligent weight fusion technology: Based on the weight distribution principle of the distance from the center of the tile image, it realizes pixel-level smooth transition in the overlapping area, significantly reduces the visibility of the stitching seam, and at the same time maintains the tonal consistency of the full-frame image, avoiding the color jump problem caused by the traditional stitching method. Description of the Drawings

[0043] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objects, and advantages of the present application will become more obvious:

[0044] Figure 1 is a flowchart of a method for unifying the resolution of variable-resolution images according to the present invention;

[0045] Figure 2 is a schematic diagram of the weighted fusion process in a method for unifying the resolution of variable-resolution images according to the present invention. Detailed implementation manners

[0046] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. In addition, it should be noted that, for the convenience of description, only the parts related to the invention are shown in the drawings.

[0047] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the drawings and embodiments.

[0048] The present invention provides a method for unifying the resolution of variable-resolution images, and the method includes the following steps:

[0049] Step S1, obtaining a remote sensing image with variable resolution to be unified as an initial image, taking the center of the initial image as a starting point, and cutting out a plurality of sliced images along the edge part of the initial image based on the variation range of the resolution from the center to the edge, and ensuring that there is an overlapping area between adjacent sliced images;

[0050] Step S2, taking the sliced image corresponding to the low-resolution area as input data, inputting it into a pre-constructed and trained super-resolution reconstruction model, and obtaining a high-resolution sliced image with the same resolution as the sliced image in the center;

[0051] Step S3, performing geometric correction processing on the sliced images after resolution unification, and correcting all sliced images into a unified coordinate system;

[0052] Step S4, performing weighted fusion processing on the overlapping areas of the sliced images after geometric correction, and realizing seamless splicing with consistent tone based on the weight distribution principle of the distance from the center of the sliced image, and generating an image with unified resolution.

[0053] For a clearer description of the method for unifying the resolution of variable-resolution images of the present invention, the following will be combined with Figure 1 Each step in the embodiments of the present invention will be described in detail.

[0054] A method for unifying the resolution of variable-resolution images according to the first embodiment of the present invention includes steps S1-S4, and each step is described in detail as follows:

[0055] Step S1, obtaining a remote sensing image with variable resolution to be unified as an initial image, taking the center of the initial image as a starting point, and cutting out a plurality of sliced images along the edge part of the initial image based on the variation range of the resolution from the center to the edge, and ensuring that there is an overlapping area between adjacent sliced images;

[0056] Specifically, multiple piece images are cropped from the initial image along the edge part. The method is as follows: starting from the center of the initial image, advancing the cropping towards the edge of the initial image, and keeping the cropping path consistent with the direction of the resolution change gradient.

[0057] The size of the piece image is dynamically adjusted according to the resolution change amplitude at its location; among them, the size of the piece image in the central area is larger than that in the edge area. Specifically, the size cropping method is as follows:

[0058] ;

[0059] ;

[0060] Among them, k represents the spatial resolution of the pixel points at the edge of the cropped image (close to the center position of the whole variable-resolution remote sensing image). If the spatial resolution of this point is not an integer multiple, it is rounded to an integer; W and H respectively represent the width and height of the cropped image when the resolution is fixed at 1 meter in the variable-resolution remote sensing image; 、 respectively represent the actual width and height of the cropped image.

[0061] In this embodiment, various existing open-source software is used to crop a variable-resolution image. When cropping, it is necessary to crop according to the resolution, that is, for a variable-resolution image, the spatial resolution decreases from 1m to 4m in sequence from the middle to the edge. Therefore, when cropping, attention should be paid to the resolution of the original image to make the resolutions on both sides of the cropped piece image as close as possible. Since the resolution of the variable-resolution image changes in an exponential curve from the center to the edge, this means that in the area close to the center of the image, the change in spatial resolution is relatively slow. Therefore, the size of the image cropped from this area is relatively large.

[0062] On the contrary, as we gradually approach the edge of the image, the change speed of the spatial resolution accelerates. Correspondingly, the size of the image cropped from these areas will be relatively small. This change characteristic of the resolution requires that when processing the variable-resolution image, appropriate cropping should be carried out according to different resolution areas to ensure that the resolutions of all the piece images after the subsequent super-resolution reconstruction are consistent. At the same time, a certain overlapping area should be designed during cropping to facilitate the subsequent splicing and fusion of the images.

[0063] Step S2, using the piece image corresponding to the low-resolution area as input data and inputting it into the pre-constructed and trained super-resolution reconstruction model to obtain a high-resolution piece image with the same resolution as the piece image in the center;

[0064] The super-resolution reconstruction model includes a shallow feature extraction module, a deep feature extraction module, and a reconstruction module connected in sequence;

[0065] The shallow feature extraction module extracts primary features through single-layer convolution operations; it can efficiently extract preliminary feature information from the input image, laying a foundation for subsequent feature processing and reconstruction tasks;

[0066] The deep feature extraction module is composed of multiple cascaded residual dense blocks. Each residual dense block contains multiple cascaded residual dense attention modules. The residual dense attention module includes multiple sequentially connected convolutional layers and one spatial attention module, and adjacent residual dense attention modules are connected through layer normalization operations;

[0067] The output of the spatial attention module is added to the input of the residual dense attention module through a residual connection;

[0068] The input of the deep feature extraction module is the primary feature output by the shallow feature extraction module, and the output of the deep feature extraction module is the deep feature;

[0069] The reconstruction module fuses the primary feature and the deep feature, and after fusion, performs upsampling processing on the fused feature through the nearest neighbor interpolation method to generate the final high-resolution image.

[0070] Specifically, in this embodiment, the deep feature extraction module is composed of 10 cascaded residual dense blocks. Each residual dense block contains 3 cascaded residual dense attention modules. The residual dense attention module includes 4 sequentially connected convolutional layers and one spatial attention module, and adjacent residual dense attention modules are connected through layer normalization operations;

[0071] In this embodiment, the batch normalization (BatchNormalization, BN) layer in the traditional dense block (DenseBlock) is removed because the BN layer often introduces artifacts, thus affecting the quality of the reconstructed image. Instead, layer normalization (Layer Normalization, LN) is introduced between two residual dense attention modules to reduce the computational complexity of the model and improve the training speed. In this way, while retaining the powerful feature extraction ability of the residual dense block, the RDAB further optimizes the performance of the model. In addition, the residual dense block (RDB) originally composed of 5 convolutional layers is improved to be composed of 4 convolutional layers and one spatial attention module. Each convolutional layer uses a 3×3 convolutional kernel, with a stride of 1 and a padding of 1, and is followed by a ReLU activation function after each layer.

[0072] The spatial attention module is mainly implemented based on depthwise separable convolution, which consists of depthwise convolution (DWConv) and pointwise convolution (PWConv). First, depthwise convolution performs convolution operations independently on each channel, retaining rich local features. Subsequently, pointwise convolution combines the features of all channels into the output feature map. This design not only reduces the computational complexity but also improves the overall performance of the network, making the model more efficient and stable when processing inputs of different resolutions:

[0073] The execution process of the spatial attention module includes:

[0074] The number of channels is doubled through 1×1 convolution and the features are divided into two parts. One part serves as the gating feature, which is processed by 7×7 depthwise separable convolution to capture large-scale spatial information; the other part serves as the main feature. The main feature is multiplied by the processed gating feature to achieve feature modulation. Finally, the number of feature channels is compressed back to the original size through 1×1 convolution.

[0075] Depthwise convolution and pointwise convolution are sequentially performed on the input feature map to generate a spatial attention weight map; the spatial attention weight map is multiplied by the input feature map channel by channel to output the weighted feature map :

[0076] ;

[0077] wherein, is depthwise convolution, is pointwise convolution, A is the gating feature, and B is the main feature.

[0078] wherein, the magnification ratio of the upsampling layer is dynamically determined by the resolution difference between the edge region and the central region, ensuring that the output image resolution is consistent with the sub-satellite point spatial resolution.

[0079] Step S3, perform geometric correction processing on the segmented images with unified resolution, and correct all segmented images into a unified coordinate system:

[0080] Step S31, construct a rigorous imaging geometric model of the satellite remote sensing image;

[0081] Specifically, the satellite remote sensing image uses the rigorous imaging geometric model to calculate the object space position of the image point, and its geometric positioning model:

[0082] ;

[0083] In the formula, is the ground point coordinate, denotes the position of the projection center measured by GPS at the imaging moment, and is a scale factor. represents the transformation matrix from the satellite body coordinate system to the J2000 coordinate system. represents the transformation matrix from the J2000 coordinate system to the WGS84 coordinate system. represents the transformation matrix from the sensor coordinate system to the satellite body coordinate system. is the coordinate of the object space vector in the sensor coordinates. is the projection center offset vector. is the focal plane distortion. is the focal length of the camera x , y is the coordinate of the image point on the sensor imaging plane

[0084] After the pixel coordinates (I, J) of the target point in the tiled image are obtained, the scan line number is determined from the pixel coordinates, and the imaging moment is obtained based on the auxiliary data. Further, the position and attitude at the imaging moment of the scan line are obtained by the satellite position and attitude interpolation method.

[0085] Step S32: Based on each tiled image, determine whether the elevation value of the target point can be obtained. If so, jump to step S33; otherwise, jump to step S34.

[0086] Step S33: Based on the rigorous imaging geometric model, given the tiled image coordinates, camera parameters, satellite orbit parameters, and the elevation value of the target point, solve the model equation to obtain the object space plane coordinates of each target point in the tiled image as the corrected coordinates ;

[0087] Step S34: Use the local average elevation of the area covered by the tiled image as the initial value, calculate the initial longitude and latitude, interpolate the elevation value corresponding to the initial longitude and latitude in the DEM data, and substitute it into the imaging geometric positioning model for iterative calculation until the results of two adjacent iterations are less than the preset threshold, then obtain the longitude, latitude, and elevation value of the tiled image , and jump to step S33.

[0088] After forward and inverse coordinate transformation and gray resampling, all images are corrected to a unified coordinate system. The prerequisite for mosaicking is to correct the images to a unified coordinate system, and then ensure that the contrast of the mosaicked images is consistent, the tones are similar, and there are no obvious seams. Therefore, the operation of step S4 is performed.

[0089] Step S4: Perform weighted fusion processing on the overlapping areas of the geometrically corrected tiled images, and achieve seamless mosaicking with consistent tones based on the weight distribution principle of the distance from the center of the tiled image, and generate an image with a unified resolution.

[0090] Since all variable-resolution images are derived from the same parent variable-resolution image, the color difference between slices is relatively small. During the splicing process, this solution uses the method of weighted summation to process the overlapping area to ensure the coherence of the tone. Specifically, for any pixel at a point within the overlapping area, the weight assignment principle is: the closer to the center point of an image, the higher the weight of this image at this pixel point; conversely, an image farther away is assigned a lower weight. Through this weighted fusion technology, the tone of the spliced image is finally made consistent.

[0091] Refer to Figure 2 the example. If we want to determine the pixel value at point P, first determine that the rectangular frame of the spliced image is composed of the right edge and the lower edge of image A, and the left edge and the upper edge of image B. It is known that the distance from point P to the right edge of image A is and the distance to the upper edge is ; the distance to the left edge of image B is and the distance to the upper edge is .

[0092] Specifically, perform weighted fusion processing on the overlapping area of the geometrically corrected sliced images. The method is as follows:

[0093] Step S41, for any pixel at a point P within the overlapping area, dynamically assign weights according to its distance from the centers of adjacent sliced images;

[0094] Step S42, calculate the pixel at point P based on the weight assignment:

[0095] ;

[0096] Among them, the two sliced images corresponding to the overlapping area where point P is located are respectively defined as image A and image B; is the pixel value of image A at point P, is the pixel value of image B at point P, is the weight of image A for the pixel at point P, is the weight of image B for the pixel at point P.

[0097] The weights and the weight , their calculation methods are:

[0098] ； ;

[0099] Among them, 、 is the distance from point P to the right edge and the lower edge of image A, 、 The distances from point P to the left edge and the upper edge of image B.

[0100] After super-resolution reconstruction, geometric correction, and image mosaicking, a variable-resolution image with the same resolution as the sub-satellite point resolution can be obtained.

[0101] Although the steps are described in the above order in the above embodiments, those skilled in the art can understand that, in order to achieve the effects of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reversed order, and these simple changes are all within the protection scope of the present invention.

[0102] A variable-resolution image resolution unification system according to the second embodiment of the present invention is based on a variable-resolution image resolution unification method according to the first embodiment. The system includes:

[0103] An image cropping module configured to obtain a remote sensing image with a variable resolution to be unified as an initial image. Starting from the center of the initial image, based on the variation range of the resolution from the center to the edge, multiple partial images are cropped from the edge part of the initial image, and it is ensured that there is an overlapping area between adjacent partial images;

[0104] A resolution unification module configured to use the partial images corresponding to the low-resolution areas as input data and input them into a pre-constructed and trained super-resolution reconstruction model to obtain high-resolution partial images with the same resolution as the partial images at the center;

[0105] A coordinate unification module configured to perform geometric correction processing on the partial images after resolution unification and correct all partial images into a unified coordinate system;

[0106] A mosaicking module configured to perform weighted fusion processing on the overlapping areas of the geometrically corrected partial images, and achieve seamless mosaicking with consistent tone based on the weight distribution principle of the distances from the centers of the partial images, and generate a unified-resolution image. Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes and related descriptions of the above-described system can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0107] It should be noted that, for the variable-resolution image resolution unification system provided in the above embodiments, only the division of the above functional modules is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further split into multiple sub-modules to complete all or part of the functions described above. For the names of the modules and steps involved in the embodiments of the present invention, they are only used to distinguish each module or step and are not regarded as an improper limitation of the present invention.

[0108] An electronic device according to a third embodiment of the present invention includes:

[0109] At least one processor; and

[0110] A memory communicatively connected to at least one of the processors; wherein,

[0111] The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned variable-resolution image resolution unification method.

[0112] A computer-readable storage medium according to a fourth embodiment of the present invention, the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned variable-resolution image resolution unification method.

[0113] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes and related descriptions of the above-mentioned storage device and processing device can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0114] Those skilled in the art should be able to realize that the modules and method steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. The programs corresponding to the software modules and method steps can be placed in a random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in the form of electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0115] The terms "first", "second", etc. are used to distinguish similar objects and are not used to describe or indicate a particular order or sequence.

[0116] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to those processes, methods, articles, or apparatus / devices.

[0117] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easily understood by those skilled in the art that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

Claims

1. A method for unifying the resolution of variable-resolution images, characterized in that, The method includes the following steps: Step S1: Obtain the remote sensing image to be unified with variable resolution as the initial image. Taking the center of the initial image as the starting point, based on the change amplitude of the resolution from the center to the edge, cut out multiple piecewise images along the edge part of the initial image, and ensure that there is an overlapping area between adjacent piecewise images; Step S2: Use the piecewise image corresponding to the low-resolution area as input data and input it into the pre-constructed and trained super-resolution reconstruction model to obtain a high-resolution piecewise image with the same resolution as the piecewise image at the center; Step S3: Perform geometric correction processing on the piecewise images with unified resolution and correct all piecewise images into a unified coordinate system: Step S31: Construct a rigorous imaging geometric model of the satellite remote sensing image; Step S32: Based on each piecewise image, determine whether the elevation value of the target point can be obtained. If so, jump to Step S33; otherwise, jump to Step S34; Step S33: Based on the rigorous imaging geometric model, given the piecewise image coordinates, camera parameters, satellite orbit parameters, and the elevation value of the target point, solve the model equation to obtain the object plane coordinates of each target point in the piecewise image as the corrected coordinates; Step S34: Use the local average elevation of the area covered by the piecewise image as the initial value, calculate the initial longitude and latitude, interpolate the elevation value corresponding to the initial longitude and latitude in the DEM data, and substitute it into the imaging geometric positioning model for iterative calculation to obtain the longitude, latitude, and elevation values of the piecewise image, and then jump to Step S33; Step S4: Perform weighted fusion processing on the overlapping areas of the geometrically corrected piecewise images, and achieve seamless stitching with consistent tone based on the weight distribution principle of the distance from the center of the piecewise image to generate an image with unified resolution.

2. The method for unifying the resolution of a variable-resolution image according to claim 1, characterized in that, The method of cutting out multiple piecewise images along the edge part of the initial image is as follows: Taking the center of the initial image as the starting point, advance the cutting towards the edge of the initial image, and keep the cutting path consistent with the resolution change gradient direction.

3. A method for unifying the resolution of variable-resolution images according to claim 1, characterized in that The size of the piecewise image is dynamically adjusted according to the change amplitude of the resolution at its location; among them, the size of the piecewise image in the central area is larger than that of the piecewise image in the edge area.

4. A method for unifying the resolution of variable-resolution images according to claim 1, characterized in that, The super-resolution reconstruction model includes a shallow feature extraction module, a deep feature extraction module, and a reconstruction module connected in sequence; The shallow feature extraction module extracts primary features through single-layer convolution operations; The deep feature extraction module is composed of multiple cascaded residual dense blocks. Each residual dense block contains multiple cascaded residual dense attention modules. The residual dense attention module includes multiple sequentially connected convolutional layers and one spatial attention module, and adjacent residual dense attention modules are connected through layer normalization operations; The output of the spatial attention module is added to the input of the residual dense attention module through a residual connection; The input of the deep feature extraction module is the primary features output by the shallow feature extraction module, and the output of the deep feature extraction module is deep features; The reconstruction module fuses the primary features and the deep features, and performs upsampling processing on the fused features through the nearest neighbor interpolation method to generate the final high-resolution image.

5. A method for unifying the resolution of variable-resolution images according to claim 4, characterized in that, The execution process of the spatial attention module includes: Performing depth convolution and pointwise convolution on the input feature map in sequence to generate a spatial attention weight map; multiplying the spatial attention weight map and the input feature map channel by channel, and outputting the weighted feature map.

6. A method for unifying the resolution of variable-resolution images according to claim 1, characterized in that, Substituting into the imaging geometric positioning model for iterative calculation, until the results of two adjacent iterations are less than the preset threshold, the longitude, latitude, and elevation values of the patch image are obtained.

7. A method for unifying the resolution of variable-resolution images according to claim 1, characterized in that, Performing weighted fusion processing on the overlapping regions of the geometrically corrected patch images, and the method is as follows: Step S41, for the pixel of any point P in the overlapping region, dynamically allocate weights according to its distance from the center of the adjacent patch image; Step S42, calculating the pixel of point P based on the weight allocation: ； Among them, the two piecewise images corresponding to the overlapping area where point P is located are respectively defined as image A and image B; is the pixel value of image A at the position of point P, is the pixel value of image B at the position of point P, is the weight of image A for the pixel of point P, is the weight of image B for the pixel of point P.

8. A method for unifying the resolution of variable-resolution images according to claim 7, characterized in that, The weight , and its calculation method is as follows: ； Among them, , are the distances from point P to the right and lower edges of image A, , are the distances from point P to the left and upper edges of image B.

9. A method for unifying the resolution of variable-resolution images according to claim 8, characterized in that, The weight , and its calculation method is as follows: ； Wherein, and are the distances from point P to the right and lower edges of image A, and are the distances from point P to the left and upper edges of image B.

Images