Unmanned aerial vehicle SAR image splicing method and system
By registering the preprocessing and local projection transformation model of the drone SAR image, and matching it with the directional logarithmic gradient histogram descriptor, the geometric distortion and noise problems in the stitching of the drone SAR image are solved, and rapid automatic Stitching and high-quality image generation.
Patent Information
- Application Number
- CN202411033083.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-22
AI Technical Summary
The existing technology cannot effectively solve the problem of rapid and automatic stitching of multiple drone SAR images, especially due to the characteristics of large geometric distortion, strong coherent spot noise and high resolution, the optical image stitching method cannot be adapted, and the lack of suitable for drone SAR images Splicing technology.
By preprocessing the drone SAR images, including image correction and matching, registration is performed using distance-weighted local projection transformation models, and weighted fusion of image intensity is performed in the overlapping area of the stitched image, combined with directional logarithmic gradient histogram description Sub- and Gaussian kernel functions perform template matching to achieve rapid and automatic stitching of images.
Eliminate the overall geometric distortion of the multi-view drone SAR image, suppress the influence of local geometric distortion and coherent spots of the image, and realize the rapid automatic splicing of the drone SAR image. The generated image artifacts are obviously eliminated, and the tone is natural, which improves the The robustness of splicing.
Smart Images

Figure CN119006274A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of SAR image data processing, and in particular to a method and system for stitching unmanned aerial vehicle SAR images. Background Art
[0002] With the popularization of civil unmanned aerial vehicles (UAVs) and the lightweight development of synthetic aperture radar (SAR) systems, it is possible to carry out surveying and mapping tasks by carrying SAR systems on UAVs. UAVs have the advantages of high flight flexibility, low manufacturing cost and easy use, and SAR systems have the ability to image all day and all weather. UAV SAR systems can have the advantages of both and have great application potential in emergency surveying and environmental monitoring.
[0003] Due to the low flying altitude of drones and the limited viewing angle of radar, a single drone SAR image usually covers a small area and cannot meet the needs of large-scale mapping tasks. Therefore, it is necessary to quickly stitch a series of overlapping multi-view drone SAR images into a SAR image that completely covers the target area. Multi-view drone SAR images have the characteristics of large geometric distortion, strong coherent speckle noise, and high resolution, making the automatic stitching of drone SAR images a very challenging problem. Existing optical image stitching methods cannot adapt to these characteristics, and there is currently no drone SAR image stitching technology published. It is urgent to solve the problem of fast and automatic stitching of multiple drone SAR images. Summary of the invention
[0004] To this end, the present invention provides a UAV SAR image stitching method and system to solve the problem of fast and automatic stitching of multiple UAV SAR images.
[0005] According to the design scheme provided by the present invention, on the one hand, a method for stitching UAV SAR images is provided, comprising:
[0006] Preprocessing the UAV SAR image to obtain a SAR image with pixel coordinates, wherein the preprocessing includes image correction;
[0007] Match the preprocessed SAR images to obtain the images to be stitched;
[0008] The distance-weighted local projection transformation model is used to register the images to be stitched, and the image intensity is weightedly fused in the overlapping area of the registered stitched images to obtain a complete SAR image covering the target area.
[0009] As the UAV SAR image stitching method of the present invention, further, the UAV SAR image is preprocessed, including:
[0010] According to the average elevation value of the target area and the image point coordinates of the four corner points of the SAR image, the range Doppler geometric imaging model of the UAV SAR image is used to obtain the object space coordinates corresponding to the four corner points, and the circumscribed rectangle is used as the object space range corresponding to the SAR image;
[0011] The object space range is divided into equally spaced regular grids. The three-dimensional coordinates of the grid points are obtained according to the object space coordinates of the grid points and the public digital elevation model. The image point coordinates of the grid points are obtained according to the range Doppler geometric imaging model.
[0012] As the UAV SAR image stitching method of the present invention, further, matching the preprocessed SAR images to obtain the image to be stitched includes:
[0013] Obtain the overlapping area of the pre-processed SAR images by comparing geographic information;
[0014] Selecting a reference image, and setting grid points at intervals in the overlapping area of the reference image;
[0015] A template window is selected on the reference image with the grid point as the center, and the oriented logarithmic gradient histogram descriptor of the grid point is extracted using the template window, and the position of the grid point on the image to be matched is estimated according to the geographic information of the SAR image, wherein the template window has overlapping image blocks;
[0016] In the local search window of the estimated position, the oriented logarithmic gradient histogram descriptor is extracted pixel by pixel, and the correlation is calculated with the oriented logarithmic gradient histogram descriptor of the grid point to obtain the matching measure;
[0017] The maximum value of the matching measure within the search window is obtained through a fast search matching strategy, and the SAR image matching points are obtained by fitting near the maximum value.
[0018] As the UAV SAR image stitching method of the present invention, further, the directional logarithmic gradient histogram descriptor of the grid point is extracted using the template window, comprising:
[0019] Perform logarithmic operations on SAR images to convert multiplicative noise into additive noise;
[0020] Use the Gaussian kernel function to weight the Sobel operator convolution window, expand the Sobel operator to multiple scales, and obtain the gradient of the image along the specified direction;
[0021] Obtain the image block gradient distribution histogram centered on each pixel in the image block, and use the L2 norm to perform normalization to obtain the descriptor vector of the image block;
[0022] The descriptor vectors of all image patches are concatenated to form a oriented log-gradient histogram descriptor.
[0023] As the UAV SAR image stitching method of the present invention, further, correlation calculation is performed with the directional logarithmic gradient histogram descriptor of the grid points and a matching measure is obtained, including:
[0024] The similarity between the template window and the search window of each channel in the image matching process is calculated based on the image block descriptor in the SAR image;
[0025] The channels whose similarity maximum value does not appear at the search window boundary are determined as valid channels, and the sum of the valid channels is obtained;
[0026] The matching position of the template window in the search window is obtained based on the sum of the valid channels.
[0027] As the UAV SAR image stitching method of the present invention, further, obtaining the maximum value of the matching measure in the search window through a fast search matching strategy includes:
[0028] Set the search window size and search grid points, scale the search window according to the reduction factor, extract the oriented logarithmic gradient histogram descriptor on the reduced search grid points and calculate the matching measure, and update the search window center according to the location of the maximum matching measure;
[0029] Centered on the updated search window, the search window size is adjusted, the oriented logarithmic gradient histogram descriptor is extracted pixel by pixel within the search window, and the matching measure is calculated to fit the image matching points within the region where the maximum matching measure is located.
[0030] As the UAV SAR image stitching method of the present invention, further, the distance-weighted local projection transformation model is used to align the images to be stitched, and it also includes:
[0031] The image to be stitched is divided into grids with a certain interval. The weighted function is used to smoothly change the movement of the image. The corresponding local model is generated at the center of each grid. All pixels in the grid use the same local model for matching mapping.
[0032] In another aspect, the present invention further provides a UAV SAR image stitching system, comprising: an image processing module, an image matching module and an image stitching module, wherein:
[0033] An image processing module is used to preprocess the UAV SAR image to obtain a SAR image with pixel coordinates, wherein the preprocessing includes image correction;
[0034] An image matching module is used to match the preprocessed SAR images to obtain the image to be stitched;
[0035] The image stitching module is used to register the images to be stitched using a distance-weighted local projection transformation model, and to perform weighted fusion of image intensities in the overlapping areas of the registered stitched images to obtain a complete SAR image covering the target area.
[0036] Beneficial effects of the present invention:
[0037] The present invention eliminates the overall geometric distortion of multi-view UAV SAR images through image correction, performs template matching based on logarithmic gradient selective correlation measure, suppresses the influence of local geometric distortion and coherent speckle of the image, and obtains reliable matching point pairs; accurately aligns images through a mobile direct linear transformation model and adopts a weighted fusion algorithm to perform image intensity fusion, so as to realize rapid generation of large-scene UAV SAR images with significantly eliminated artifacts and natural color transition from multiple UAV SAR images, realize rapid and automatic splicing of UAV SAR images, and improve the robustness of SAR image splicing. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 This is a schematic diagram of the UAV SAR image stitching process in the embodiment;
[0039] Figure 2 This is an example of the UAV SAR image correction result in the embodiment;
[0040] Figure 3 This is an example of the UAV SAR image matching result in the embodiment;
[0041] Figure 4 This is an example of the UAV SAR image registration result in the embodiment;
[0042] Figure 5 This is an example of the stitching result of the UAV SAR image in the embodiment. DETAILED DESCRIPTION
[0043] In order to make the purpose, technical solutions and advantages of the present invention clearer and more understandable, the present invention is further described in detail below in conjunction with the accompanying drawings and technical solutions.
[0044] Embodiments of the present invention, see Figure 1 As shown, a UAV SAR image stitching method is provided, comprising:
[0045] S101, preprocessing the UAV SAR image to obtain a SAR image with pixel coordinates, wherein the preprocessing includes image correction.
[0046] For the collected UAV SAR images, they can be corrected by an indirect method based on the UAV SAR system parameters, Position and Orientation System (POS) data and the external public digital elevation model (Digital Elevation Model) and the geometric construction model of the UAV SAR image.
[0047] S102: Match the preprocessed SAR images to obtain images to be stitched.
[0048] For two overlapping calibrated UAV SAR images, the template matching method based on logarithmic gradient selective correlation measure can be used for image matching.
[0049] S103, using a distance-weighted local projection transformation model to register the images to be stitched, and performing weighted fusion of image intensities on the overlapped areas of the registered stitched images to obtain a complete SAR image covering the target area.
[0050] For two matching UAV SAR images, the Moving Direct Linear Transformation (MDLT) model can be used to accurately align the two images, and then the weighted fusion algorithm can be used to perform image intensity fusion.
[0051] Among them, the preprocessing of UAV SAR images can be designed to include the following contents:
[0052] According to the average elevation value of the target area and the image point coordinates of the four corner points of the SAR image, the range Doppler geometric imaging model of the UAV SAR image is used to obtain the object space coordinates corresponding to the four corner points, and the circumscribed rectangle is used as the object space range corresponding to the SAR image;
[0053] The object space range is divided into equally spaced regular grids. The three-dimensional coordinates of the grid points are obtained according to the object space coordinates of the grid points and the public digital elevation model. The image point coordinates of the grid points are obtained according to the range Doppler geometric imaging model.
[0054] According to the average elevation value of the test area and the image point coordinates of the four corner points of the image, the RD model is used to solve the object space coordinates corresponding to the four corner points, and the circumscribed rectangle of the solution result is calculated, which is used as the object space range corresponding to the SAR image. The parameters of the RD model are obtained from the system parameters and POS data. The object space area is divided into equally spaced regular grids, and the grid interval is set to the sampling interval of the corrected image. The image points of the corrected image correspond one-to-one to the object space grid points. According to the object space coordinates of the grid points, the elevation values are interpolated on the public DEM to obtain the three-dimensional coordinates of the grid points, and the image point coordinates of the grid points are solved by the RD model. The obtained image point coordinates are generally not integers, and the grayscale value of the point needs to be interpolated on the original image, and the grayscale value is assigned to the corresponding image point of the corrected image.
[0055] Among them, matching the preprocessed SAR images to obtain the images to be stitched can be designed to include:
[0056] Obtain the overlapping area of the pre-processed SAR images by comparing geographic information;
[0057] Selecting a reference image, and setting grid points at intervals in the overlapping area of the reference image;
[0058] A template window is selected on the reference image with the grid point as the center, and the oriented logarithmic gradient histogram descriptor of the grid point is extracted using the template window, and the position of the grid point on the image to be matched is estimated according to the geographic information of the SAR image, wherein the template window has overlapping image blocks;
[0059] In the local search window of the estimated position, the oriented logarithmic gradient histogram descriptor is extracted pixel by pixel, and the correlation is calculated with the oriented logarithmic gradient histogram descriptor of the grid point to obtain the matching measure;
[0060] The maximum value of the matching measure within the search window is obtained through a fast search matching strategy, and the SAR image matching points are obtained by fitting near the maximum value.
[0061] For the two corrected UAV SAR images, since the geographic information is known, the overlapping area of the two images can be obtained by comparing the geographic information. In the overlapping area of the reference image, the grid points for matching are evenly set at a certain interval. The logarithmic gradient of the image is calculated, and a template window of a certain size is selected with the grid point as the center on the reference image to extract the Histogram of Orientated Log Gradient (HOLG) descriptor; the position of the grid point on the image to be matched is estimated according to the geographic information of the two images; within the local search window of the estimated position, the HOLG descriptor is extracted pixel by pixel, and the selective correlation is calculated with the HOLG descriptor of the grid point to construct the matching measure HOLGSC; a fast search matching strategy is used to find the maximum value of HOLGSC in the search window, and fit near the maximum value to obtain the exact matching point.
[0062] The oriented logarithmic gradient histogram descriptor of the grid points is extracted using the template window and can be designed to include:
[0063] Perform logarithmic operations on SAR images to convert multiplicative noise into additive noise;
[0064] Use the Gaussian kernel function to weight the Sobel operator convolution window, expand the Sobel operator to multiple scales, and obtain the gradient of the image along the specified direction;
[0065] Obtain the image block gradient distribution histogram centered on each pixel in the image block, and use the L2 norm to perform normalization to obtain the descriptor vector of the image block;
[0066] The descriptor vectors of all image patches are concatenated to form a oriented log-gradient histogram descriptor.
[0067] Perform logarithmic operation on the SAR image to convert the multiplicative noise into additive noise, that is,
[0068] J(x,y)=log(I(x,y)+γ) (1)
[0069] Where I represents the input image and J represents the transformed image. γ is a small value to avoid zero input logarithmic function.
[0070] Use the Gaussian kernel function to weight the Sobel operator convolution window and expand the Sobel operator to multiple scales. The gradient of the direction can be calculated as
[0071]
[0072] Among them, f1 and f2 are a pair of Gaussian weighted filters, that is,
[0073]
[0074] Where α is the standard deviation of the Gaussian kernel function, also known as the scale parameter. The side length of the filter window is 2α+1. Then the gradient can be calculated as
[0075]
[0076] The gradient distribution histogram is generated by the image blocks with a certain overlap in the template window to form a block descriptor, and then all block descriptors form a HOLG descriptor. The generation of the HOLG descriptor can be summarized as including the following three steps:
[0077] a) Calculate the log gradient of the image.
[0078] b) Extract block descriptors. For each pixel in the image, set the block size centered on the pixel to N. b ×N b Pixel, calculate the gradient distribution histogram in the block, the number of direction intervals of the histogram is N a ; Normalize the descriptor using the L2 norm to obtain the descriptor vector of the block.
[0079] c) Generate HOLG descriptor. Select a template window of a certain size on the image, connect the descriptor vectors of all blocks in the template window, and the overlap between blocks is β to obtain the descriptor vector V of the template window.
[0080] Among them, calculating the correlation with the oriented logarithmic gradient histogram descriptor of the grid points and obtaining the matching measure may include:
[0081] The similarity between the template window and the search window of each channel in the image matching process is calculated based on the image block descriptor in the SAR image;
[0082] The channels whose similarity maximum value does not appear at the search window boundary are determined as valid channels, and the sum of all valid channels is obtained;
[0083] The matching position of the template window in the search window is obtained based on the sum of the valid channels.
[0084] The HOLG descriptor is a combination of multiple block descriptors. A part of these block descriptors can be selected as valid descriptors, and the other block descriptors can be selected as invalid descriptors. The valid part corresponds to the local geometric invariant features, while the invalid part corresponds to the local geometric change features. Let V = [v1, v2, ..., v N ] and U=[u1,u2,…,u N] are the HOLG descriptors of the reference image and the image to be matched, respectively. N is the number of blocks in the template window. The NCC value of the i-th block represents the similarity between the template window and the search window in the i-th channel, which is defined as
[0085]
[0086] Among them, v i and u i are the i-th block descriptors in V and U respectively; and They are v i and u i The average value of, i = 1, 2, ..., N; (x v ,y v ) is the position of the template window, (x u ,y u ) is the estimated position of the search window, (m,n) is the offset position of the search window, and r is the radius of the search window.
[0087] A valid block usually corresponds to a significant peak in the similarity measure of the corresponding channel. Therefore, the condition for selecting a channel is: if the maximum value of the i-th similarity measure channel appears at the boundary of the search window, the channel is considered to be an invalid channel; otherwise, the channel is considered to be a valid channel, that is,
[0088]
[0089] The sum of all valid channels is
[0090]
[0091] The matching position of the template window in the search window can be obtained, that is,
[0092]
[0093] Specifically, the maximum value of the matching measure within the search window is obtained through a fast search matching strategy, which can be designed to include:
[0094] Set the search window size and search grid points, scale the search window according to the reduction factor, extract the oriented logarithmic gradient histogram descriptor on the reduced search grid points and calculate the matching measure, and update the search window center according to the location of the maximum matching measure;
[0095] Centered on the updated search window, the search window size is adjusted, the oriented logarithmic gradient histogram descriptor is extracted pixel by pixel within the search window, and the matching measure is calculated to fit the image matching points within the region where the maximum matching measure is located.
[0096] A two-step search and matching strategy from coarse to fine is adopted. The search window size is set to (2r+1)×(2r+1), and the one-dimensional search grid points are (-r,-r+1,…,0,1,…,r). In the coarse search and matching stage, the search window is first scaled by a certain reduction factor s, so the one-dimensional search grid points are (-r,-r+s,…,0,s,…,r), and the search window size is reduced to (2r / s+1)×(2r / s+1). The descriptors are extracted and HOLG is calculated only on the reduced search grid points. SC , find HOLG SC The center of the updated search window is the maximum value. In the fine search matching stage, the search window size is set to (2s+3)×(2s+3) with the updated search window as the center, and the one-dimensional search grid points are (-s-1,-s,…,0,1,…,s+1). Extract the descriptor pixel by pixel in the search window and calculate the HOLG SC , find HOLG SC The maximum value of is obtained, and the exact matching point is obtained by fitting near the maximum value.
[0097] For the two matching UAV SAR images, the MDLT model is first used to accurately register the two images, and then the weighted fusion algorithm is used to perform image intensity fusion.
[0098] The MDLT model is a distance-weighted local projection transformation model that can be used to accurately register two images with complex geometric distortions. The weighting function changes smoothly as the image point moves on the image, thereby generating a transformation model that can flexibly adapt to the matching data. * , the weight function used to estimate the local transformation model at this point is
[0099]
[0100] Among them, σ is a scale parameter, x i is the coordinate of the i-th matching point in the reference image I, and γ is a small constant used to limit the extrapolation range of the weight. The above formula can be closer to x i Image point x * Higher weights are assigned, so the estimated local model H * Can better take care of x * The surrounding local structure.
[0101] When x * When moving continuously on I, H * will change smoothly, but each pixel estimates H * It is very time-consuming, so the image can be divided into grids with a certain interval, and an H is calculated for each grid center point. * , all pixels in the grid use the same H* Mapping is performed to improve computational efficiency.
[0102] For two UAV SAR images that are precisely aligned in geometry, weighted fusion of image intensity can be performed in the overlap area. Exponential weighted fusion is performed on the center line of the overlap area, and the one-dimensional exponential weighting function W(x) is:
[0103]
[0104] Where b is a point on the center line; k is a weighting coefficient, k>0; w is half of the weighted fusion range in the direction perpendicular to the center line. Assuming that the weight is less than 0.1, the influence of the weighting term on the fusion result can be ignored. k=-log(0.2) / w 2 .
[0105] Furthermore, based on the above method, an embodiment of the present invention also provides a UAV SAR image stitching system, comprising: an image processing module, an image matching module and an image stitching module, wherein:
[0106] An image processing module is used to preprocess the UAV SAR image to obtain a SAR image with pixel coordinates, wherein the preprocessing includes image correction;
[0107] An image matching module is used to match the preprocessed SAR images to obtain the image to be stitched;
[0108] The image stitching module is used to register the images to be stitched using a distance-weighted local projection transformation model, and to perform weighted fusion of image intensities in the overlapping areas of the registered stitched images to obtain a complete SAR image covering the target area.
[0109] In order to verify the effectiveness of this solution, the following is a further explanation based on experimental data:
[0110] The UAV SAR images in the experiment were acquired by a fixed-wing UAV. Figure 1 The steps in this case solution are image correction, image matching and image stitching in sequence, and the results are as follows Figures 2 to 5 shown.
[0111] Figure 2 This is an example of the correction result of a single UAV SAR image. Figure 3 This is an example of the matching results of two UAV SAR images. Figure 4 This is an example of the registration results of two UAV SAR images. Figure 5 This is an example of stitching results of 12 UAV SAR images.
[0112] The above experimental data show that this solution can quickly and automatically generate large-format UAV SAR images with significantly eliminated artifacts and natural color transitions, which can be applied to emergency mapping and environmental monitoring and other fields.
[0113] Unless otherwise specifically stated, the relative steps, numerical expressions and values of the components and steps set forth in these embodiments do not limit the scope of the present invention.
[0114] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0115] The units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person of ordinary skill in the art may use different methods to implement the described functions for each specific application, but such implementation is not considered to be beyond the scope of the present invention.
[0116] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits, and accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software function modules. The present invention is not limited to any specific form of combination of hardware and software.
[0117] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims
1. A UAV SAR image stitching method, characterized in that: Include: Preprocessing the UAV SAR image to obtain a SAR image with pixel coordinates, wherein the preprocessing includes image correction; Match the preprocessed SAR images to obtain the images to be stitched; The distance-weighted local projection transformation model is used to register the images to be stitched, and the image intensity is weightedly fused in the overlapping area of the registered stitched images to obtain a complete SAR image covering the target area.
2. The UAV SAR image stitching method according to claim 1, characterized in that: Preprocessing of UAV SAR images, including: According to the average elevation value of the target area and the image point coordinates of the four corner points of the SAR image, the range Doppler geometric imaging model of the UAV SAR image is used to obtain the object space coordinates corresponding to the four corner points, and the circumscribed rectangle is used as the object space range corresponding to the SAR image; The object space range is divided into equally spaced regular grids. The three-dimensional coordinates of the grid points are obtained according to the object space coordinates of the grid points and the public digital elevation model. The image point coordinates of the grid points are obtained according to the range Doppler geometric imaging model.
3. The UAV SAR image stitching method according to claim 1, characterized in that: Match the preprocessed SAR images to obtain the images to be stitched, including: Obtain the overlapping area of the pre-processed SAR images by comparing geographic information; Selecting a reference image, and setting grid points at intervals in the overlapping area of the reference image; A template window is selected on the reference image with the grid point as the center, and the oriented logarithmic gradient histogram descriptor of the grid point is extracted using the template window, and the position of the grid point on the image to be matched is estimated according to the geographic information of the SAR image, wherein the template window has overlapping image blocks; In the local search window of the estimated position, the oriented logarithmic gradient histogram descriptor is extracted pixel by pixel, and the correlation is calculated with the oriented logarithmic gradient histogram descriptor of the grid point to obtain the matching measure; The maximum value of the matching measure within the search window is obtained through a fast search matching strategy, and the SAR image matching points are obtained by fitting near the maximum value.
4. The UAV SAR image stitching method according to claim 3, characterized in that: The oriented logarithmic gradient histogram descriptor of the grid points is extracted using the template window, including: Perform logarithmic operations on SAR images to convert multiplicative noise into additive noise; Use the Gaussian kernel function to weight the Sobel operator convolution window, expand the Sobel operator to multiple scales, and obtain the gradient of the image along the specified direction; Obtain the image block gradient distribution histogram centered on each pixel in the image block, and use the L2 norm to perform normalization to obtain the descriptor vector of the image block; The descriptor vectors of all image patches are concatenated to form a oriented log-gradient histogram descriptor.
5. The UAV SAR image stitching method according to claim 3, characterized in that: Calculate the correlation with the oriented logarithmic gradient histogram descriptor of the grid points and obtain the matching measure, including: The similarity between the template window and the search window of each channel in the image matching process is calculated based on the image block descriptor in the SAR image; The channels whose similarity maximum value does not appear at the search window boundary are determined as valid channels, and the sum of the valid channels is obtained; The matching position of the template window in the search window is obtained based on the sum of the valid channels.
6. The UAV SAR image stitching method according to claim 3, characterized in that: The maximum value of the matching measure within the search window is obtained through the fast search matching strategy, including: Set the search window size and search grid points, scale the search window according to the reduction factor, extract the oriented logarithmic gradient histogram descriptor on the reduced search grid points and calculate the matching measure, and update the search window center according to the location of the maximum matching measure; Centered on the updated search window, the search window size is adjusted, the oriented logarithmic gradient histogram descriptor is extracted pixel by pixel within the search window, and the matching measure is calculated to fit the image matching points within the region where the maximum matching measure is located.
7. The UAV SAR image stitching method according to claim 1, characterized in that: The distance-weighted local projection transformation model is used to align the stitched images, which also includes: The image to be stitched is divided into grids with a certain interval. The weighted function is used to smoothly change the movement of the image. The corresponding local model is generated at the center of each grid. All pixels in the grid use the same local model for matching mapping.
8. A UAV SAR image stitching system, characterized in that: It includes: image processing module, image matching module and image stitching module, among which, An image processing module is used to preprocess the UAV SAR image to obtain a SAR image with pixel coordinates, wherein the preprocessing includes image correction; An image matching module is used to match the preprocessed SAR images to obtain the image to be stitched; The image stitching module is used to register the images to be stitched using a distance-weighted local projection transformation model, and to perform weighted fusion of image intensities in the overlapping areas of the registered stitched images to obtain a complete SAR image covering the target area.
9. An electronic device, characterized in that: include: at least one processor, and a memory coupled to the at least one processor; The memory stores a computer program, and the computer program can be executed by the at least one processor to implement the method according to any one of claims 1 to 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, the method according to any one of claims 1 to 7 can be implemented.
