Optical remote sensing image uniform and dense homonymy point matching method, system and device based on dense optical flow and storage medium

Abstract

The invention discloses an optical remote sensing image uniform and dense homonymy point matching method, system and device based on dense optical flow, and a storage medium, belongs to the technical field of satellite remote sensing image processing, and solves the problem that sufficient and uniformly distributed homonymy points can not be identified in various complex scenes in the prior art. The invention aims to solve the technical problems in the prior art and to provide the dense optical flow-based uniform and dense homonymy point matching method for the optical remote sensing image. Confirming a remote sensing image overlapping region, and dividing the remote sensing image overlapping region into a plurality of remote sensing image blocks; constructing a displacement field resolving model and a Gaussian image pyramid to obtain a displacement field of the remote sensing image overlapping region; and obtaining a pre-matched homonymy point pair set, distributing the pre-matched homonymy point pair set to the corresponding remote sensing image overlapping region image blocks, and screening to obtain a final matched homonymy point set. According to the optical remote sensing image uniform and dense homonymy point matching method based on the dense optical flow, sufficient and uniformly distributed homonymy points can be identified in various complex scenes, and the positioning precision is high.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite remote sensing image processing, and in particular to a method, system, device and storage medium for uniform and dense homonymous point matching of optical remote sensing images based on dense optical flow. Background Art

[0002] Common approaches to matching homonymous points in high-resolution optical remote sensing satellite imagery can be broadly categorized into deep learning methods and traditional algorithms based on a "feature extraction, description, matching, and false match rejection" strategy. Data-driven deep learning methods are highly dependent on training samples, resulting in poor accuracy for existing unsupervised learning neural networks. Supervised learning neural networks, on the other hand, require large, highly manually annotated datasets. Furthermore, homonymous point annotation in optical remote sensing imagery spanning large areas, long temporal spans, and diverse imaging conditions presents challenges such as a large labeling load, high recognition difficulty, and overly monotonous scene coverage. Traditional algorithms based on an "extraction, description, matching" strategy perform well in areas with prominent features and stable structural information. However, they often struggle to identify homonymous points in regions with weak texture, complex local distortion, or significant grayscale appearance differences. Furthermore, due to the limitations of multi-stage matching strategies, even if sufficient identical features are extracted between two images, these algorithms may still fail to provide sufficient homonymous points if the feature description algorithm cannot accurately construct a discriminative feature space or if the pre-defined consistency model deviates significantly from the actual spatial mapping relationship. In addition, existing traditional algorithms mostly target prominent geometric structural features in images. For images of complex landform types such as mixed mountainous and urban areas, their recognition results often show a high degree of spatial aggregation effect.

[0003] In the prior art, Chinese patent document CN103279935A discloses a "Method and System for Super-Resolution Reconstruction of Thermal Infrared Remote Sensing Images Based on the MAP Algorithm." This method acquires a sequence of thermal infrared remote sensing images, comprising at least two frames. It employs a high-precision automatic registration method based on corner point features, utilizing automatic extraction and matching of corner points to achieve registration. It also utilizes the MAP algorithm to achieve super-resolution reconstruction of the sequence of images, and proposes an adaptive method for selecting the threshold of the edge penalty function for the potential function parameters of the Gibbs model. Finally, it performs application-oriented quality evaluation on the reconstructed target resolution image. However, this technical solution utilizes automatic extraction and matching of corner points to achieve registration, making it difficult to identify homonymous points in areas with weak texture, complex local distortion, or large differences in grayscale appearance. Furthermore, due to the multi-stage matching strategy, its recognition results often exhibit a high degree of spatial clustering for complex terrain types, such as mixed mountainous and urban areas.

[0004] In summary, the existing technology lacks a technical problem of a uniformly dense homonymous point matching method for optical remote sensing images based on dense optical flow, which can identify sufficient and evenly distributed homonymous points in various complex scenes with high positioning accuracy. Summary of the Invention

[0005] The present invention solves the technical problem that the existing technology lacks a method for matching uniformly dense homonymous points in optical remote sensing images based on dense optical flow, which can identify sufficient and evenly distributed homonymous points in various complex scenes and has high positioning accuracy.

[0006] The present invention provides a method for matching uniformly dense homonymous points in optical remote sensing images based on dense optical flow, comprising the following steps: Step 1: confirming a remote sensing image overlap area based on geographic coordinate information of a plurality of remote sensing images, dividing the remote sensing image overlap area into a plurality of remote sensing image blocks, and obtaining remote sensing image overlap area image blocks; Step 2: construct a displacement field solution model and a Gaussian image pyramid to obtain the displacement field of the overlapping area of the remote sensing image; Step 3: Design a sparse homonymous point sampling strategy to obtain a set of pre-matched homonymous point pairs; Step 4: Assign the same-name points in the pre-matched same-name point pair set to the corresponding remote sensing image overlapping area image blocks, screen out the correctly matched same-name points in the remote sensing image overlapping area image blocks, and use the set of correctly matched same-name points as the final matched same-name point set.

[0007] Furthermore, in an embodiment of the present invention, the size of the remote sensing image block in step 1 is no larger than 800 pixels*800 pixels.

[0008] Furthermore, in the embodiment of the present invention, the step 2 of constructing the displacement field solution model and the Gaussian image pyramid to obtain the displacement field of the image block in the overlapping area of the remote sensing image is specifically as follows: The size of the image blocks in the overlapping area of the remote sensing images is expanded. Based on the spatial grayscale information and gradient structure information of the image blocks in the overlapping area of the remote sensing images, a displacement field smoothing constraint is introduced, and a displacement field solution model is constructed to obtain the primary displacement field of the image blocks in the overlapping area of the remote sensing images. After resampling and sampling parameter correction at the top layer of the Gaussian image pyramid, the primary displacement field of the image blocks in the overlapping area of the remote sensing images is used as the initial value for calculating the displacement field of the next layer in the Gaussian image pyramid. The calculation result of the bottom layer of the Gaussian image pyramid is used as the expanded displacement field of the image blocks in the overlapping area of the remote sensing images. The expanded part of the image blocks in the overlapping area of the remote sensing images is eliminated, and the middle part of the image blocks in the overlapping area of the remote sensing images is used as the displacement field of the image blocks in the overlapping area of the remote sensing images.

[0009] Furthermore, in the embodiment of the present invention, the sparse homonymous point sampling strategy in step 3 is used to obtain a set of pre-matched homonymous point pairs, specifically: A set of candidate corner points of a remote sensing image is extracted. Based on the minimum distance between adjacent points, the overlapping area of the remote sensing image is divided into multiple adjacent grids. The corner point closest to the center of each adjacent grid is screened out from the set of candidate corner points. Based on the displacement field of the overlapping area of the remote sensing image, the same-name points corresponding to the corner point closest to the center of the adjacent grid are obtained, and a set of pre-matched same-name point pairs is obtained.

[0010] Furthermore, in an embodiment of the present invention, the extraction of a set of candidate corner points of a remote sensing image is specifically as follows: The corner points of the remote sensing image are extracted, and the displacement field of the overlapping area of the remote sensing image is uniformly sampled to obtain the supplementary point set of the remote sensing image. The corner points of the remote sensing image and the supplementary point set of the remote sensing image are combined to obtain the candidate corner point set of the remote sensing image.

[0011] The present invention provides a system for matching uniformly dense homonymous points in optical remote sensing images based on dense optical flow, comprising the following modules: A division module, which confirms the overlapping area of remote sensing images based on the geographic coordinate information of multiple remote sensing images, divides the overlapping area of remote sensing images into multiple remote sensing image blocks, and obtains image blocks of the overlapping area of remote sensing images; The construction module constructs the displacement field solution model and Gaussian image pyramid respectively to obtain the displacement field of the overlapping area of the remote sensing image; Matching module, designs sparse homonymous point sampling strategy and obtains pre-matched homonymous point pair sets; The screening module assigns the same-name points in the pre-matched same-name point pair set to the corresponding image blocks in the overlapping area of the remote sensing image, screens out the correctly matched same-name points in the image blocks in the overlapping area of the remote sensing image, and uses the set of correctly matched same-name points as the final matched same-name point set.

[0012] An electronic device according to the present invention comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is configured to implement any of the above-mentioned methods for uniformly dense matching of homonymous points in optical remote sensing images based on dense optical flow when executing the program stored in the memory.

[0013] The present invention provides a computer-readable storage medium having a computer program stored therein. When the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for uniformly dense homonymous point matching in optical remote sensing images based on dense optical flow.

[0014] The present invention solves the technical problem that the existing technology lacks a method for matching uniformly dense homonymous points in optical remote sensing images based on dense optical flow, which can identify sufficient and evenly distributed homonymous points in various complex scenes and has high positioning accuracy. Specific beneficial effects include: This paper addresses the problem of identifying homonymous points in high-spatial-resolution optical remote sensing images and proposes a method for uniformly and densely matching homonymous points in optical remote sensing images based on dense optical flow. The overlapping areas of the remote sensing images are roughly located based on the geographic coordinate information of multiple remote sensing images. The spatial grayscale information and gradient structure information of the remote sensing images are combined to construct a displacement field calculation model to solve the spatial correspondence between the two remote sensing images pixel by pixel. A Gaussian image pyramid is used to address the large displacement problem between remote sensing images. Corner point structures in remote sensing images are detected through corner point extraction, which can more accurately describe the spatial position and geometric structure characteristics of remote sensing images. Corner points are uniformly sampled through grid sampling to improve the uniformity of the corner point set. The corresponding homonymous points are then determined based on the displacement field of the overlapping areas of the remote sensing images. A random sampling consistency algorithm based on a local affine model is used to eliminate mismatched homonymous point pairs. This addresses the potential for severe local geometric inconsistencies and overfitting of the displacement field in remote sensing images of different phases, thereby achieving uniformly and densely matching high-precision homonymous points in complex grayscale and geometric distortion scenarios. This uniform and dense matching method for optical remote sensing images based on dense optical flow can extract sufficient, evenly distributed and dense high-precision homonymous points in various complex image scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a schematic diagram of the results of extracting the same-name points from the urban construction area image in the first embodiment; Figure 2 This is a schematic diagram of the results of extracting the same-name points from the mountain image described in the first embodiment; Figure 3 This is a schematic diagram of the results of extracting homonymous points in a plain area according to the first embodiment; Figure 4 This is a schematic diagram of the extraction results of homonymous points in a mixed terrain area described in Implementation Method 1. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe various embodiments of the present invention in conjunction with the accompanying drawings. The embodiments described with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0017] Implementation 1. This implementation describes a method for uniformly dense matching of homonymous points in optical remote sensing images based on dense optical flow, comprising the following steps: Step 1: confirming a remote sensing image overlap area based on geographic coordinate information of a plurality of remote sensing images, dividing the remote sensing image overlap area into a plurality of remote sensing image blocks, and obtaining remote sensing image overlap area image blocks; Step 2: construct a displacement field solution model and a Gaussian image pyramid to obtain the displacement field of the overlapping area of the remote sensing image; Step 3: Design a sparse homonymous point sampling strategy to obtain a set of pre-matched homonymous point pairs; Step 4: Assign the same-name points in the pre-matched same-name point pair set to the corresponding remote sensing image overlapping area image blocks, screen out the correctly matched same-name points in the remote sensing image overlapping area image blocks, and use the set of correctly matched same-name points as the final matched same-name point set.

[0018] In this embodiment, the size of the remote sensing image block in step 1 is no larger than 800 pixels*800 pixels.

[0019] In this embodiment, the displacement field calculation model and Gaussian image pyramid are constructed in step 2 to obtain the displacement field of the image block in the overlapping area of the remote sensing image, specifically: The size of the image blocks in the overlapping area of the remote sensing images is expanded. Based on the spatial grayscale information and gradient structure information of the image blocks in the overlapping area of the remote sensing images, a displacement field smoothing constraint is introduced, and a displacement field solution model is constructed to obtain the primary displacement field of the image blocks in the overlapping area of the remote sensing images. After resampling and sampling parameter correction at the top layer of the Gaussian image pyramid, the primary displacement field of the image blocks in the overlapping area of the remote sensing images is used as the initial value for calculating the displacement field of the next layer in the Gaussian image pyramid. The calculation result of the bottom layer of the Gaussian image pyramid is used as the expanded displacement field of the image blocks in the overlapping area of the remote sensing images. The expanded part of the image blocks in the overlapping area of the remote sensing images is eliminated, and the middle part of the image blocks in the overlapping area of the remote sensing images is used as the displacement field of the image blocks in the overlapping area of the remote sensing images.

[0020] In this embodiment, the sparse homonymous point sampling strategy in step 3 is used to obtain a set of pre-matched homonymous point pairs, specifically: A set of candidate corner points of a remote sensing image is extracted. Based on the minimum distance between adjacent points, the overlapping area of the remote sensing image is divided into multiple adjacent grids. The corner point closest to the center of each adjacent grid is screened out from the set of candidate corner points. Based on the displacement field of the overlapping area of the remote sensing image, the same-name points corresponding to the corner point closest to the center of the adjacent grid are obtained, and a set of pre-matched same-name point pairs is obtained.

[0021] In this embodiment, the extraction of a set of candidate corner points of a remote sensing image is specifically as follows: The corner points of the remote sensing image are extracted, and the displacement field of the overlapping area of the remote sensing image is uniformly sampled to obtain the supplementary point set of the remote sensing image. The corner points of the remote sensing image and the supplementary point set of the remote sensing image are combined to obtain the candidate corner point set of the remote sensing image.

[0022] The existing technology has the technical problem of lacking a method for matching uniformly dense homonymous points in optical remote sensing images based on dense optical flow, which can identify sufficient and evenly distributed homonymous points in various complex scenes and has high positioning accuracy.

[0023] To solve the above technical problems, this embodiment provides a method for matching uniformly dense homonymous points in optical remote sensing images based on dense optical flow, which solves the problem of identifying homonymous points between optical remote sensing images of different phases and the same spatial resolution. This embodiment replaces remote sensing images of different phases with Img1 and Img2, respectively, and specifically includes the following steps: Step 1: confirming a remote sensing image overlap area based on geographic coordinate information of a plurality of remote sensing images, dividing the remote sensing image overlap area into a plurality of remote sensing image blocks, and obtaining remote sensing image overlap area image blocks; Step 1 is to roughly locate the overlapping areas of remote sensing images based on geographic coordinate information and to implement a blocking strategy for the overlapping areas of remote sensing images. To reduce the amount of computation and improve the efficiency of the homonymous point recognition method, it is first necessary to determine the approximate overlapping range of the remote sensing images based on the geographic coordinate information accompanying the remote sensing images when they are released. Using geographic coordinate information such as the geographic coordinates of the starting point in the upper left corner of the remote sensing image and the spatial resolution, a mapping relationship between remote sensing image pixels and geographic coordinates can be established, namely: ; (1) ; (2) in, is the geographic coordinate of the starting point of the upper left corner of the image, is the image spatial resolution parameter, For pixels The image plane coordinates, For pixels The geographical coordinates of .

[0024] Formula (1) can be used to calculate the boundary positions of the remote sensing image in the east, west, south and north directions, that is, the four boundaries of the remote sensing image. Among them, the four boundaries of Img1 are , the range of Img2 is At this point, Img1 and Img2 are unified into the same geographic coordinate system, so the overlapping area of the remote sensing images can be calculated. If the remote sensing images do not have overlapping areas, the recognition process ends immediately, and the set of identified points with the same name is empty. If the remote sensing images do have overlapping areas, the subsequent steps are continued.

[0025] Since it is difficult to achieve high-precision positioning without sensor control, it is impossible to accurately determine the position of the remote sensing image overlap area by relying solely on the geographic coordinate information of the remote sensing image. In order to reduce the impact of the non-remote sensing image overlap area on subsequent methods, it is necessary to shrink the remote sensing image overlap area to a certain size, and the range of the remote sensing image overlap area is obtained as follows: , the overlapping area of remote sensing images can be mapped to the grid coordinate system of Img1 through formula (2) to obtain the overlapping area of Img1.

[0026] If the overlap area is too large, the efficiency and accuracy of dense optical flow calculations will be severely affected. Therefore, the overlap area of Img1 is evenly divided into multiple remote sensing image blocks, each of which is no larger than 800 pixels by 800 pixels. When solving the displacement field, the image block is used as the calculation unit.

[0027] Step 2: construct a displacement field solution model and a Gaussian image pyramid to obtain the displacement field of the overlapping area of the remote sensing image; The size of the image blocks in the overlapping area of the remote sensing images is expanded. Based on the spatial grayscale information and gradient structure information of the image blocks in the overlapping area of the remote sensing images, a displacement field smoothing constraint is introduced, and a displacement field solution model is constructed to obtain the primary displacement field of the image blocks in the overlapping area of the remote sensing images. After resampling and sampling parameter correction at the top layer of the Gaussian image pyramid, the primary displacement field of the image blocks in the overlapping area of the remote sensing images is used as the initial value for calculating the displacement field of the next layer in the Gaussian image pyramid. The calculation result of the bottom layer of the Gaussian image pyramid is used as the expanded displacement field of the image blocks in the overlapping area of the remote sensing images. The expanded part of the image blocks in the overlapping area of the remote sensing images is eliminated, and the middle part of the image blocks in the overlapping area of the remote sensing images is used as the displacement field of the image blocks in the overlapping area of the remote sensing images.

[0028] Step 2 is to construct a displacement field solution model based on dense optical flow. Affected by factors such as imaging conditions and seasonal conditions, remote sensing images of different phases are prone to obvious nonlinear differences in spatial grayscale information. It is difficult to ensure the stability of a uniform and dense matching method of homonymous points in optical remote sensing images based on dense optical flow by constructing a displacement field solution model based only on the assumption of constant illumination. Therefore, the method used in this embodiment is to combine the spatial grayscale information and gradient structure information of remote sensing images, introduce displacement field smoothing constraints, and jointly construct a displacement field solution model. The displacement field solution model expression is as follows: ; (3) ; (4) ; (5) ; (6) ; (7) in, is the gradient structure weight set a priori, is the smoothness constraint parameter, To avoid the penalty term being an infinite decimal of 0, The general value is 0.001. = is the displacement field to be determined, is any pixel position in the image, is the gradient operator, is the spatial grayscale information, is the gradient structure information, is the displacement field smoothness constraint. By solving the minimization energy function The primary displacement field of the image blocks in the overlapping area of the remote sensing images is obtained.

[0029] However, the primary displacement field of the image blocks in the overlapping area of the remote sensing images calculated by the displacement field solution model has a large displacement problem, which leads to inaccurate matching of the same-name points. To address the possible large displacement problem of Img1 and Img2, a multi-level solution is performed using the pyramid image strategy, specifically: First, manually predefine the remote sensing image downsampling parameters and the maximum allowable displacement , the default values are 0.75 and 30 respectively, and then the number of pyramid layers is calculated according to the following formula , and construct a Gaussian image pyramid: ; (8) Assume that the size of the lowest layer remote sensing image is , for the Layer pyramid image, whose size is The displacement field is calculated step by step from the top layer of the pyramid. The result of the current layer is resampled and the sampling parameters are corrected before it is used as the initial value for the displacement field calculation of the next layer. In addition, considering that the accuracy of the displacement field usually decreases significantly in the edge area, before solving each image block, it is necessary to expand it to a certain size as the bound. In this implementation, the default value of the bound is 30 pixels. After the calculation is completed, the expanded part is removed and only the middle part is retained as the displacement field of the current image block. The calculation result of the bottom image is the displacement field of the image block in the overlapping area of the remote sensing images. After calculating all image blocks separately, the displacement field of the overlapping area of the remote sensing images Img1 and Img2 is obtained.

[0030] However, using the displacement field of the overlapping area of remote sensing images to match the remote sensing image homonymous points will cause problems of image radiation and geometric distortion. Therefore, a sparse homonymous point sampling strategy is proposed to solve the problems of image radiation and geometric distortion.

[0031] Step 3: Design a sparse homonymous point sampling strategy to obtain a set of pre-matched homonymous point pairs; A set of candidate corner points of a remote sensing image is extracted. Based on the minimum distance between adjacent points, the overlapping area of the remote sensing image is divided into multiple adjacent grids. The corner point closest to the center of each adjacent grid is screened out from the set of candidate corner points. Based on the displacement field of the overlapping area of the remote sensing image, the same-name points corresponding to the corner point closest to the center of the adjacent grid are obtained, and a set of pre-matched same-name point pairs is obtained.

[0032] Step 3 designs a sparse same-name point sampling strategy based on the corner point structure of remote sensing images. According to the grayscale information and gradient structure information of the overlapping area of remote sensing images, the spatial position correspondence between the same-name points of the two remote sensing images is established and sparse same-name point sampling is performed.

[0033] Corner point structure features are typically highly significant and stable. On the one hand, they facilitate the correct matching of homonymous points through manual discrimination and other a posteriori methods. On the other hand, corner point structures are generally robust to image radiometric and geometric distortions, such as those at road intersections and building vertices. Compared to randomly selected points, corner points can more accurately describe the spatial location and geometric structure of remote sensing images.

[0034] In this embodiment, the Harris operator is used to extract the corner points of Img1 block by block, and the corner point set is obtained. Considering that corner point structures are prone to being undetectable or over-aggregated in mountainous areas and urban construction areas, the displacement field of the overlapping area of remote sensing images is uniformly sampled to obtain a supplementary corner point set. , fill in points in weak texture areas or other areas where corner points are difficult to extract as much as possible, where the sampling radius It is a priori set value, and the default value is 30. and Together they form a set of candidate corner points .

[0035] In order to improve the uniformity of the point set, it is necessary to further screen the set of corner points to be selected. Define the minimum distance between adjacent points , divide the overlapping area of remote sensing images into multiple For grids of different sizes, grids with insufficient size at the edges can be merged with adjacent grids or no special processing is performed. Among the multiple corner points in the grid, retain the corner point closest to the grid center and delete the other corner points. The coordinates of the corresponding points with the same name can be obtained through the displacement field of the overlapping area of the remote sensing image in step 2.

[0036] In Img1, any corner point The image coordinates are , displacement field At this corner The pixel value at is expressed as the displacement Therefore, Img2 Corresponding points of the same name The image coordinates are .

[0037] By using the sparse sampling strategy of the same-name points, we can filter out the candidate points that are too close as much as possible, and get the pre-matched same-name point pair set. .in, For the For points of the same name, and are their coordinates in the pixel coordinate systems of Img1 and Img2 respectively.

[0038] Step 4: assign the same-name points in the pre-matched same-name point pair set to the corresponding remote sensing image overlap area image blocks, screen out the correct matching same-name points in the remote sensing image overlap area image blocks, and use the set of correct matching same-name points as the final matching same-name point set; Step 4 is a method for eliminating mismatched homonymous points based on a local affine model. Due to objective factors such as imaging angle and changes in ground elevation, remote sensing images from different time phases may exhibit severe local geometric inconsistencies. For example, farmland, viaducts, and buildings in the same area may experience geometric deformations of varying degrees and directions. The displacement field calculated in step 2 in the overlapping areas of remote sensing images is prone to overfitting, i.e., incorrect stretching. Furthermore, shadows and fog can also affect the accuracy of the calculated displacement field in the overlapping areas of remote sensing images. Therefore, further screening of the pre-matched homonymous point pairs in step 3 is necessary to eliminate mismatched point pairs.

[0039] This implementation uses a random sampling consensus algorithm to pre-match the set of homonymous point pairs. The number of iterations and the model deviation parameters can be manually adjusted, with the default values being 1000 and 3.0, respectively. Considering the large size of remote sensing images and the large differences in local geometric distortion in different regions, the global model estimated using partial points cannot accurately express the mapping relationship between each point, making it impossible to determine the correctness of the matching of homonymous points.

[0040] Therefore, this embodiment will Each point is assigned to the image block of the overlapping area of the remote sensing image divided in step 1. Each image block of the overlapping area of the remote sensing image is independently screened using RANSAC. Finally, the screening results of the image blocks of the overlapping area of the remote sensing image are integrated into the same set to obtain the final matching point set with the same name.

[0041] This embodiment proposes a method for matching uniformly dense homonymous points in optical remote sensing images based on dense optical flow. By combining the spatial grayscale information and gradient structure information of the remote sensing images and introducing smoothness constraints, the spatial correspondence between the overlapping areas of remote sensing images of different phases is established pixel by pixel, and the displacement field of the overlapping areas of the remote sensing images is obtained. Then, based on the displacement field, the sparse homonymous point sampling strategy and the local affine model proposed in this embodiment are used for sampling and screening, and finally a high-precision final matching homonymous point set is obtained.

[0042] Compared to the commonly used "feature extraction, description, and matching" approach, this implementation can identify sufficient and evenly distributed keypoints in a variety of complex scenarios, with high positioning accuracy. Compared to deep learning algorithms, this implementation requires less labor and material resources, and exhibits excellent universality and robustness to complex image nonlinear grayscale mapping and local geometric distortion.

[0043] In order to better illustrate the method for uniformly dense matching of same-name points in optical remote sensing images based on dense optical flow described in this embodiment, the following examples are used for detailed description: Example 1. This embodiment establishes the spatial correspondence of the overlapping area of the remote sensing images pixel by pixel based on the spatial grayscale information and gradient structure information of the joint remote sensing images, obtains the displacement field of the overlapping area of the remote sensing images, and then further screens and filters out the homonymous points and mismatched points through the sparse homonymous point sampling strategy, and finally obtains a sufficient number of homonymous points that are evenly distributed.

[0044] Four groups of satellite remote sensing images collected by the "Jilin-1" high-resolution satellite were used as experimental targets. Each group of remote sensing images has undergone pre-processing steps such as orthorectification, full-color multispectral fusion and cropping, and are all raster images in standard TIF format. According to the observation overlapping area scenes, the four groups of satellite remote sensing images are divided into: "mountainous group", "urban group", "plain group" and "complex terrain group", among which "complex terrain group" refers to complex mixed terrain areas including mountains, urban areas and waters. According to the different coverage scenes, it can be divided into mountainous remote sensing images, urban construction area remote sensing images, plain remote sensing images and mixed terrain remote sensing images. The resolution, size, number of bands and pixel grayscale value of the remote sensing images have no effect on the effect of this embodiment. For the convenience of processing and description, the experimental data are uniformly processed: the spatial resolution is 0.75 meters, the size is uniformly cropped to 5000*5000, the number of bands is 4, and the pixel grayscale value is .

[0045] This embodiment uses satellite remote sensing images collected by the Jilin-1 high-resolution satellite for testing, and can achieve large-scale, automated, sufficient and evenly distributed homonymous point recognition. For remote sensing images in different scenes, this embodiment can match sufficient and evenly distributed homonymous points with high precision. The specific effects are as follows: Figure 1-4This example uses default parameters for experiments.

[0046] Implementation 2. This implementation describes a dense optical flow-based uniform and dense homonymous point matching system for optical remote sensing images, comprising the following modules: A division module, which confirms the overlapping area of remote sensing images based on the geographic coordinate information of multiple remote sensing images, divides the overlapping area of remote sensing images into multiple remote sensing image blocks, and obtains image blocks of the overlapping area of remote sensing images; The construction module constructs the displacement field solution model and Gaussian image pyramid respectively to obtain the displacement field of the overlapping area of the remote sensing image; Matching module, designs sparse homonymous point sampling strategy and obtains pre-matched homonymous point pair sets; The screening module assigns the same-name points in the pre-matched same-name point pair set to the corresponding image blocks in the overlapping area of the remote sensing image, screens out the correctly matched same-name points in the image blocks in the overlapping area of the remote sensing image, and uses the set of correctly matched same-name points as the final matched same-name point set.

[0047] Implementation three. An electronic device described in this implementation includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is configured to implement the method for uniformly dense matching of same-name points in optical remote sensing images based on dense optical flow as described in the first embodiment when executing the program stored in the memory.

[0048] Implementation method 4. The computer-readable storage medium described in this implementation method stores a computer program, and when the computer program is executed by the processor, it implements the method for uniformly dense same-name point matching of optical remote sensing images based on dense optical flow described in implementation method 1.

[0049] The above is a detailed introduction to the method, system, device and storage medium for uniform and dense matching of same-name points in optical remote sensing images based on dense optical flow proposed in the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for matching uniformly dense homonymous points in optical remote sensing images based on dense optical flow, characterized in that: The following steps are involved: Step 1: confirming a remote sensing image overlap area based on geographic coordinate information of a plurality of remote sensing images, dividing the remote sensing image overlap area into a plurality of remote sensing image blocks, and obtaining remote sensing image overlap area image blocks; Step 2: construct a displacement field solution model and a Gaussian image pyramid to obtain the displacement field of the overlapping area of the remote sensing image; Step 3: Design a sparse homonymous point sampling strategy to obtain a set of pre-matched homonymous point pairs; Step 4: Assign the same-name points in the pre-matched same-name point pair set to the corresponding remote sensing image overlapping area image blocks, screen out the correctly matched same-name points in the remote sensing image overlapping area image blocks, and use the set of correctly matched same-name points as the final matched same-name point set.

2. The method for uniformly dense homonymous point matching in optical remote sensing images based on dense optical flow according to claim 1, characterized in that: The size of the remote sensing image block in step 1 is no larger than 800 pixels * 800 pixels.

3. The method for uniformly dense homonymous point matching in optical remote sensing images based on dense optical flow according to claim 1, characterized in that: The displacement field calculation model and Gaussian image pyramid are constructed in step 2 to obtain the displacement field of the image blocks in the overlapping area of the remote sensing image, specifically: The size of the image blocks in the overlapping area of the remote sensing images is expanded. Based on the spatial grayscale information and gradient structure information of the image blocks in the overlapping area of the remote sensing images, a displacement field smoothing constraint is introduced, and a displacement field solution model is constructed to obtain the primary displacement field of the image blocks in the overlapping area of the remote sensing images. After resampling and sampling parameter correction at the top layer of the Gaussian image pyramid, the primary displacement field of the image blocks in the overlapping area of the remote sensing images is used as the initial value for calculating the displacement field of the next layer in the Gaussian image pyramid. The calculation result of the bottom layer of the Gaussian image pyramid is used as the expanded displacement field of the image blocks in the overlapping area of the remote sensing images. The expanded part of the image blocks in the overlapping area of the remote sensing images is eliminated, and the middle part of the image blocks in the overlapping area of the remote sensing images is used as the displacement field of the image blocks in the overlapping area of the remote sensing images.

4. The method for uniformly dense homonymous point matching in optical remote sensing images based on dense optical flow according to claim 1, characterized in that: The sparse homonymous point sampling strategy in step 3 is used to obtain a set of pre-matched homonymous point pairs, specifically: A set of candidate corner points of a remote sensing image is extracted. Based on the minimum distance between adjacent points, the overlapping area of the remote sensing image is divided into multiple adjacent grids. The corner point closest to the center of each adjacent grid is screened out from the set of candidate corner points. Based on the displacement field of the overlapping area of the remote sensing image, the same-name points corresponding to the corner point closest to the center of the adjacent grid are obtained, and a set of pre-matched same-name point pairs is obtained.

5. The method for uniformly dense homonymous point matching in optical remote sensing images based on dense optical flow according to claim 4, characterized in that: The extraction of a set of candidate corner points of a remote sensing image is specifically as follows: The corner points of the remote sensing image are extracted, and the displacement field of the overlapping area of the remote sensing image is uniformly sampled to obtain the supplementary point set of the remote sensing image. The corner points of the remote sensing image and the supplementary point set of the remote sensing image are combined to obtain the candidate corner point set of the remote sensing image.

6. A uniform and dense homonymous point matching system for optical remote sensing images based on dense optical flow, characterized by: Includes the following modules: A division module, which confirms the overlapping area of remote sensing images based on the geographic coordinate information of multiple remote sensing images, divides the overlapping area of remote sensing images into multiple remote sensing image blocks, and obtains image blocks of the overlapping area of remote sensing images; The construction module constructs the displacement field solution model and Gaussian image pyramid respectively to obtain the displacement field of the overlapping area of the remote sensing image; Matching module, designs sparse homonymous point sampling strategy and obtains pre-matched homonymous point pair sets; The screening module assigns the same-name points in the pre-matched same-name point pair set to the corresponding image blocks in the overlapping area of the remote sensing image, screens out the correctly matched same-name points in the image blocks in the overlapping area of the remote sensing image, and uses the set of correctly matched same-name points as the final matched same-name point set.

7. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is configured to implement the method for uniformly dense matching of same-name points in optical remote sensing images based on dense optical flow as described in any one of claims 1 to 5 when executing the program stored in the memory.

8. 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 a processor, the method for uniformly dense homonymous point matching in optical remote sensing images based on dense optical flow according to any one of claims 1 to 5 is implemented.

Images