Method and apparatus for determining change regions in remote sensing images based on triangular mesh dense matching

Abstract

This invention provides a method and apparatus for determining change regions in remote sensing images based on dense matching of triangular meshes. The method includes: acquiring previous and later remote sensing images to be processed; performing feature point matching on the previous and later remote sensing images to obtain an initial set of matching points between them; constructing an irregular triangular mesh based on the initial set of matching points, and performing dense matching on the pixels inside each triangle in the irregular triangular mesh to obtain a target set of densely matched points between the previous and later remote sensing images; and determining the change region between the previous and later remote sensing images based on the target set of densely matched points. This invention is applicable not only to multi-band images but also to single-band images, and possesses high stability and robustness, enabling automated batch extraction of change regions from multi-temporal remote sensing images.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image change detection technology, and in particular to a method and apparatus for determining change areas in remote sensing images based on dense triangular mesh matching. Background Technology

[0002] Currently, existing remote sensing image change detection methods are mainly divided into two categories: multi-scale segmentation-based methods and deep learning neural network-based methods. First, existing methods have stringent data requirements, necessitating automatic registration of images from different periods before and after the change, and these images often require multi-band coverage. Second, multi-scale segmentation-based methods have low processing efficiency, and the classification of land cover types requires manual intervention; the processing accuracy often depends on the degree of human intervention. Deep learning neural network-based methods require extensive pre-training of relevant samples, resulting in significant time commitment for training and parameter tuning. Third, both existing methods have poor universality; changing the data necessitates re-evaluating the parameters and effectiveness. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method and apparatus for determining the change region of remote sensing images based on dense matching of triangular meshes. It is applicable not only to multi-band images but also to single-band images, and has high stability and robustness. It can automatically extract the change region of multi-temporal remote sensing images in batches.

[0004] In a first aspect, embodiments of the present invention provide a method for determining change regions in remote sensing images based on dense triangular mesh matching, including:

[0005] Acquire the pre-processed and post-processed remote sensing images to be processed;

[0006] Feature point matching is performed on the earlier remote sensing images and the later remote sensing images to obtain an initial set of matching points between the earlier and later remote sensing images;

[0007] An irregular triangular network is constructed based on the initial set of matching points, and dense matching is performed on the pixels inside each triangle in the irregular triangular network to obtain the target dense matching point set between the earlier remote sensing image and the later remote sensing image.

[0008] Based on the set of densely matched target points, the area of ​​change between the earlier and later remote sensing images is determined.

[0009] In one implementation, feature point matching is performed on the earlier remote sensing images and the later remote sensing images to obtain an initial set of matching points between the earlier and later remote sensing images, including:

[0010] Construct an early pyramid image corresponding to the early remote sensing image and a later pyramid image corresponding to the later remote sensing image; wherein, both the early pyramid image and the later pyramid image include multiple image layers;

[0011] Multiple feature points to be matched are extracted from the aforementioned previous remote sensing images;

[0012] By using sliding windows of different sizes, each image layer of the early pyramid image and each image layer of the later pyramid image are traversed multiple times, so that each feature point to be matched is matched using multiple feature point matching algorithms in each traversal process, and the matching point set corresponding to each traversal process is obtained.

[0013] For each feature point to be matched, if the matching result of the feature point to be matched is consistent in the matching point set corresponding to each traversal process, then the matching feature point and the matching result of the matching feature point are retained to obtain the initial matching point set.

[0014] In one implementation, dense matching is performed on the pixels inside each triangle in the irregular triangular network to obtain a set of target densely matched points between the earlier remote sensing image and the later remote sensing image, including:

[0015] Dense matching is performed on the pixels inside each triangle in the irregular triangular network to obtain an initial set of densely matched points between the earlier remote sensing image and the later remote sensing image.

[0016] Determine the affine transformation error value of each densely matched point in the initial densely matched point set. If the affine transformation error value of the densely matched point is greater than a preset error threshold, then remove the densely matched point from the initial densely matched point set to obtain an intermediate densely matched point set.

[0017] Detect missing matching points in the intermediate dense matching point set, and perform nearest neighbor matching on the missing matching points to obtain the target dense matching point set between the earlier remote sensing image and the later remote sensing image.

[0018] In one implementation, detecting missing matching points in the intermediate dense matching point set includes:

[0019] The intermediate dense matching point set is divided into a dense matching point subset corresponding to each triangle in the irregular triangular network;

[0020] The process involves iterating through each triangle in the irregular triangular mesh, determining the unmatched pixels in the dense matching point subset corresponding to each triangle based on a preset density matching degree, and identifying the unmatched pixels as missing matching points.

[0021] In one implementation, nearest neighbor matching is performed on the missed matching points to obtain a set of densely matched target points between the earlier and later remote sensing images, including:

[0022] Determine the distance between the missed matching point and each successfully matched pixel, and determine multiple target pixels from each successfully matched pixel based on the distance;

[0023] Construct an affine transformation model based on each of the target pixels;

[0024] The missing matching points are matched using the affine transformation model to obtain a set of densely matched target points between the earlier and later remote sensing images.

[0025] In one implementation, determining the change region between the earlier and later remote sensing images based on the target dense matching point set includes:

[0026] The aforementioned early remote sensing images are divided into multiple image blocks;

[0027] Based on the target dense matching point set, determine the number of dense matching points contained in each image block;

[0028] The matching success rate of each image block is determined based on the total number of pixels, the number of invalid pixels, the number of dense matching points, and the preset dense matching degree of the previous remote sensing images.

[0029] If the matching success rate of the image block is less than a preset success rate threshold, then the image block is determined to be the change region between the previous remote sensing image and the later remote sensing image.

[0030] In one implementation, the matching success rate of each image block is determined based on the total number of pixels, the number of invalid pixels, the number of densely matched points, and a preset dense matching degree of the previous remote sensing image, including:

[0031] Determine the difference between the total number of pixels and the number of invalid pixels in the previous remote sensing image;

[0032] For each image block, determine the product between the number of dense matching points contained in the image block and a preset dense matching degree;

[0033] The ratio of the difference to the product is determined as the matching success rate of the image block.

[0034] Secondly, embodiments of the present invention also provide a device for determining the change region of remote sensing images based on dense triangular mesh matching, comprising:

[0035] The image acquisition module is used to acquire pre-processed and post-processed remote sensing images to be processed.

[0036] The initial matching module is used to perform feature point matching on the earlier remote sensing images and the later remote sensing images to obtain an initial set of matching points between the earlier remote sensing images and the later remote sensing images.

[0037] The dense matching module is used to construct an irregular triangular network based on the initial matching point set, and to perform dense matching on the pixels inside each triangle in the irregular triangular network to obtain the target dense matching point set between the previous remote sensing image and the later remote sensing image.

[0038] The change region determination module is used to determine the change region between the previous remote sensing image and the later remote sensing image based on the target dense matching point set.

[0039] Thirdly, embodiments of the present invention also provide an electronic device, including a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method described in any of the first aspects.

[0040] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the method described in any of the first aspects.

[0041] This invention provides a method and apparatus for determining change regions in remote sensing images based on dense matching of triangular meshes. After acquiring the previous and later remote sensing images to be processed, feature point matching is first performed on the previous and later remote sensing images to obtain an initial set of matching points between the two images. Then, an irregular triangular mesh is constructed based on the initial set of matching points, and dense matching is performed on the pixels inside each triangle in the mesh to obtain a target set of densely matched points between the previous and later remote sensing images. Finally, the change region between the previous and later remote sensing images is determined based on the target set of densely matched points. The above method utilizes the characteristic that unchanged areas can be successfully matched while changed areas cannot, and proposes a method for determining the change area of ​​remote sensing images based on dense triangular mesh matching. This method can quickly and stably determine the range of change areas in remote sensing images, providing a basis for refined automatic change detection or manual judgment. The embodiments of this invention are applicable not only to multi-band images but also to single-band images. In addition, the embodiments of this invention do not require high-precision registration processing of pre- and post-continuous remote sensing images, nor do they require human intervention or training with a large number of samples. They have high stability and robustness and can automatically extract change areas from multi-temporal remote sensing images in batches.

[0042] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0043] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0044] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0045] Figure 1 A flowchart illustrating a method for determining change regions in remote sensing images based on dense triangular mesh matching, provided in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of the sliding window principle provided in an embodiment of the present invention;

[0047] Figure 3A flowchart illustrating a HOPC algorithm provided in an embodiment of the present invention;

[0048] Figure 4 A schematic diagram of an irregular Delaunay triangulation provided in an embodiment of the present invention;

[0049] Figure 5 A flowchart illustrating another method for determining the change region of remote sensing images based on dense triangular mesh matching provided in an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of a remote sensing image change region determination device based on dense triangular mesh matching provided in an embodiment of the present invention;

[0051] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Currently, related technologies provide the following methods for detecting changes in remote sensing images:

[0054] 1. Method and System for Remote Sensing Image Change Detection: This invention discloses a method and system for remote sensing image change detection. The method includes the following steps: using vector patches corresponding to a previous time-phase remote sensing image as a reference, segmenting a subsequent time-phase remote sensing image to obtain segmented patches for the subsequent time-phase remote sensing image; extracting all or part of the spectral and textural features of the corresponding region in the previous time-phase remote sensing image from the segmented patches to construct a previous spatial vector; extracting features corresponding to the previous spatial vector features from the selected segmented patches to construct a subsequent spatial vector feature, and constructing a subsequent spatial vector based on the subsequent spatial feature vector; and determining whether the selected segmented patches have changed by comparing the subsequent spatial vector and the previous spatial vector. This method features low salt-and-pepper noise and high detection accuracy.

[0055] 2. A method and system for detecting changes in multi-temporal and multispectral remote sensing images: This invention discloses a method and system for detecting changes in multi-temporal and multispectral remote sensing images. The method first utilizes non-negative matrix factorization to fuse the change vector magnitudes of multi-temporal remote sensing images with multi-temporal spectral angle mapping maps to obtain new difference images. Then, the FCM algorithm is applied to the difference images to obtain the unary energy term of the CRF. Next, based on the image's neighborhood and the difference images, the binary energy term of the CRF is obtained. Finally, the energy of the CRF is minimized using the circular belief propagation algorithm to obtain the final change detection result. This invention can better characterize the relationships between image neighborhoods, improving the accuracy of change detection; the change detection results are more reliable and robust.

[0056] 3. A Remote Sensing Image Change Detection Method Based on Siamese Convolutional Neural Networks: This invention discloses a remote sensing image change detection method based on Siamese convolutional neural networks, relating to the field of remote sensing, and mainly addressing the problem of poor generalization of current conventional change detection methods. The method includes the following steps: acquiring multi-temporal remote sensing image data, obtaining mask images, establishing a remote sensing image change detection dataset, constructing a Siamese convolutional neural network model, training the Siamese convolutional neural network using the dataset to obtain a trained model, using the trained model to perform change detection on the preceding and following temporal images to obtain preliminary change prediction results, comparing the predicted pixel values ​​of the preliminary change prediction results with a pre-set pixel threshold, thereby classifying the preliminary change prediction results into changed area categories and non-changed area categories, and obtaining the change detection result. The method of this invention has better generalization performance and meets end-to-end processing requirements, facilitating engineering applications.

[0057] However, the aforementioned remote sensing image change detection methods all suffer from problems such as high image requirements, low efficiency, and poor universality. Based on this, the present invention provides a method and device for determining the change area of ​​remote sensing images based on dense triangular mesh matching. It is applicable not only to multi-band images but also to single-band images, and has high stability and robustness. It can automatically extract the change area of ​​multi-temporal remote sensing images in batches.

[0058] To facilitate understanding of this embodiment, a method for determining the change region of remote sensing images based on dense triangular mesh matching, as disclosed in this embodiment of the invention, will first be described in detail. (See [link to relevant documentation]). Figure 1 The diagram shows a flowchart of a method for determining change areas in remote sensing images based on dense triangular mesh matching. This method mainly includes the following steps S102 to S108:

[0059] Step S102: Obtain the pre-processing remote sensing images and post-processing remote sensing images to be processed.

[0060] Among them, the early remote sensing images and the later remote sensing images can be multi-band images or single-band images. The later remote sensing images can be understood as the remote sensing images of the early remote sensing images at the next moment.

[0061] Step S104: Perform feature point matching on the early and later remote sensing images to obtain an initial set of matching points between the early and later remote sensing images.

[0062] Initial feature point matching refers to automatically matching a certain number of corresponding points on two sets of remote sensing images. These corresponding points are also known as initial matching points (or initial matching points) for the next step of triangulation construction.

[0063] In one implementation, multiple different feature point matching algorithms can be used to perform initial feature point matching on both early and late remote sensing images to improve the accuracy of the initial matching point set.

[0064] Step S106: Construct an irregular triangular network based on the initial set of matching points, and perform dense matching on the pixels inside each triangle in the irregular triangular network to obtain a set of densely matched target points between the previous and later remote sensing images.

[0065] The irregular triangular mesh consists of multiple triangles, and the vertices of each triangle are initial matching points in the initial matching point set.

[0066] In one implementation, an initial set of matching points can be input into a preset tool, which then constructs an irregular triangular network based on the initial set of matching points. For each triangle in the irregular triangular network, feature point matching is performed on the pixels inside the triangle to achieve dense matching based on the irregular triangular network, thus obtaining an initial set of dense matching points. Then, missing matching points are detected on the initial set of dense matching points, and nearest neighbor matching is performed on the missing matching points to obtain the target set of dense matching points.

[0067] Step S108: Based on the set of densely matched target points, determine the area of ​​change between the previous and later remote sensing images.

[0068] The changing region can be a changing rectangular region.

[0069] In one implementation, taking advantage of the fact that unchanged areas can be successfully matched while changed areas cannot be successfully matched, the previous remote sensing image is divided into multiple image blocks, and the matching success rate of each image block is determined based on the set of dense matching points of the target. If the matching success rate of a certain image block is lower than a preset success rate threshold, it indicates that the image block is a changed area.

[0070] The remote sensing image change region determination method based on dense triangulation matching provided in this invention utilizes the characteristic that unchanged regions can be successfully matched while changed regions cannot. This method can quickly and stably determine the range of changed regions in remote sensing images, enabling refined automatic change detection or manual judgment. This invention is applicable not only to multi-band images but also to single-band images. Furthermore, this invention does not require high-precision registration of pre- and post-construction remote sensing images, nor does it require human intervention or training with a large number of samples. It exhibits high stability and robustness and can automatically extract changed regions from multi-temporal remote sensing images in batches.

[0071] To facilitate understanding, this embodiment of the invention provides a specific implementation of a method for determining change regions in remote sensing images based on dense triangular mesh matching.

[0072] Regarding step S104 above, this embodiment of the invention provides an implementation method for performing feature point matching on earlier and later remote sensing images to obtain an initial set of matching points between the earlier and later remote sensing images. Initial feature point matching refers to automatically matching a certain number of corresponding points on the two sets of remote sensing images for use in the next step of triangulation construction. There are many methods for automatic matching of image corresponding points, commonly including correlation coefficient matching, SIFT (Scale-invariant feature transform) matching, and the Histogram of Oriented Phase Congruency (HOPC) method, which supports automatic matching of heterogeneous images. To improve processing efficiency and matching accuracy, these three matching algorithms are usually based on the strategy of pyramid-level matching. Correlation coefficient matching and SIFT matching are applicable to automatic matching of homogeneous remote sensing images and are relatively efficient, while the phase congruency method can automatically match both homogeneous and heterogeneous remote sensing images with a high success rate, but its disadvantages are lower efficiency and more error points.

[0073] If the pre- and post-processing remote sensing images are not registered and the initial accuracy is unclear, direct point-by-point dense matching with a large search radius R would be extremely time-consuming. Therefore, this embodiment of the invention first automatically matches a certain number of corresponding points in the pre- and post-processing remote sensing images. These corresponding points can also be called initial matching points. The matching of initial matching points can use a large search radius to handle situations where the initial accuracy deviation is large. The accuracy of the initial matching points has a crucial impact on subsequent processing; therefore, the accuracy of the initial matching points should be improved as much as possible to ensure that all initial matching points are correct.

[0074] This invention combines the advantages and disadvantages of various methods. The initial matching points employ a pyramid-based correlation coefficient + phase consistency matching method. Each feature point must simultaneously satisfy both methods to be considered a successful match, thus improving the accuracy of the initial matching points. The matching parameter search radius R can be set to 200. Feature points are evenly distributed within the overlapping area of ​​the pre- and post-processing remote sensing images. The number of feature points N can be determined based on the image size. It can be proportionally set to extract 10,000 feature points per 100 million pixels; more feature points result in slower efficiency, while fewer feature points lead to a less dense triangular network.

[0075] This invention provides a matching method based on the correlation coefficient and phase consistency of a pyramid, and an implementation method for constructing an initial set of matching points, as shown in steps 1 to 4 below:

[0076] Step 1: Construct the early pyramid image corresponding to the early remote sensing image and the later pyramid image corresponding to the later remote sensing image; both the early pyramid image and the later pyramid image include multiple image layers.

[0077] In practical implementation, pyramid imagery is a structure that interprets images using multiple resolutions. Taking early remote sensing images as an example, N image layers of different resolutions are generated by multi-scale pixel sampling of the early remote sensing images. The image layer with the highest resolution is placed at the bottom, arranged in a pyramid shape, with a series of image layers of gradually decreasing pixel size above it, until the top of the pyramid contains only a single pixel, thus forming the pyramid image of the early remote sensing images. Similarly, later pyramid images corresponding to later remote sensing images can be constructed, which will not be elaborated further in this embodiment of the invention.

[0078] Optionally, a 2x2 early pyramid image of the early remote sensing image can be constructed using the above method, and a 2x2 late pyramid image of the late remote sensing image can be constructed using the above method.

[0079] Step 2: Extract multiple feature points to be matched from the previous remote sensing images.

[0080] In one implementation, the Moravec algorithm can be used to uniformly extract N feature points to be matched from previous remote sensing images.

[0081] Step 3: Using sliding windows of different sizes, traverse each image layer of the early pyramid image and each image layer of the later pyramid image multiple times. In each traversal, use multiple feature point matching algorithms to perform feature point matching for each feature point to be matched, and obtain the matching point set corresponding to each traversal.

[0082] The sliding window can include small-sized sliding windows, medium-sized sliding windows, and large-sized sliding windows. It should be noted that large, medium, and small are only relative concepts. In specific implementations, sliding windows of different sizes can be configured based on actual needs. This embodiment of the invention does not impose any restrictions on this. The feature point matching algorithm can be the correlation coefficient matching method and the phase consistency method.

[0083] Considering that a complete search based solely on the image itself, regardless of the matching method, is time-consuming, labor-intensive, and prone to errors, utilizing a pyramid image layer with varying resolutions from low to high to form a coarse-to-fine image matching scheme is currently the most commonly used and optimal matching method. Therefore, this embodiment of the invention can iteratively search for the point with the highest correlation in the corresponding location of each feature point to be matched using the pyramid matching method in subsequent remote sensing images.

[0084] For example, firstly, a small sliding window is used to match the feature points to be matched in each image layer. The matching algorithm uses the correlation coefficient matching method and the phase consistency method. When the matching results of the two algorithms are consistent, the matching result is retained, thus obtaining the set of matching points for this traversal. Similarly, a medium-sized sliding window is used to match the feature points to be matched in each image layer, thus obtaining the set of matching points for this traversal. Similarly, a large sliding window is used to match the feature points to be matched in each image layer, thus obtaining the set of matching points for this traversal.

[0085] Furthermore, this embodiment of the invention provides an implementation method for traversal using a sliding window, see [link to relevant documentation]. Figure 2The diagram illustrates a sliding window principle. A memory block f (size (2R1+1)*(2R1+1)) with an outer radius of R1 is read from a feature point PT1 in the previous image. Based on the principle that geographic coordinates are approximately the same, the pixel coordinates PT2 of this feature point in the later image are calculated. A memory block g (size (2R2+1)*(2R2+1)) with an outer radius of R2 is then drawn from PT2. R1 is called the template radius, and R2 is called the search radius. R2 must be greater than R1 so that f can slide through the searched image g via the window. Correlation comparison is used to find the pixel with the highest correlation coefficient. It can be seen that the larger the search radius R2, the slower the matching speed; the smaller the template radius, the faster the matching speed. The matching speed depends on the size of the template and the size of the search window. Furthermore, the reliability of the matching increases with the size of the search window. However, if there are significant differences between the two images to be matched, a large search window can actually reduce the matching accuracy or even lead to incorrect matching. With a search window that is too small, if the area to be matched does not have sufficient feature information, mismatches may also occur. Therefore, window size has a significant impact on matching success rate. Template size can generally be determined based on experience, but the search window size is difficult to find a universally suitable size due to varying image conditions.

[0086] In practical applications, to balance speed and accuracy, a dynamic window method is used for matching operations within the pyramid imagery. Specifically, for each pyramid layer imagery, at least three windows of increasing size are used to match the feature points to be matched. A match is considered successful only if the matching results are substantially consistent each time; otherwise, the sliding window is gradually enlarged to a predetermined limit to achieve a correct match.

[0087] Furthermore, embodiments of the present invention also provide specific logic for the correlation coefficient matching method and the phase consistency method, respectively:

[0088] (a) Correlation coefficient matching method:

[0089] The primary metric used in automatic matching is the correlation coefficient. Correlation coefficient matching uses the correlation coefficient (standardized covariance) as a similarity measure. In statistics, the correlation coefficient represents the correlation between two random variables; extended to image matching, it can be used to represent the degree of similarity between two images of the same size.

[0090]

[0091] R(X,Y) is called the correlation coefficient between two images. Here, E(X) and E(Y) are the mean gray values ​​of the two images, D(X) and D(Y) are the variances of the two images, and E(XY) is the mean of the product of corresponding points in the two images. Their definitions are the same as those in general statistical theory.

[0092] The correlation coefficient has the following properties:

[0093] ①R(X,Y)=R(Y,X);

[0094] ②|R(X,Y)|≤1;

[0095] ③ The necessary and sufficient condition for |R(X,Y)|=1 is that the images X and Y are linearly correlated with each other.

[0096] As can be seen, the correlation coefficient R(X,Y)| represents the degree of similarity of the linear relationship between images X and Y. The closer the correlation coefficient is to 1 or -1, the more obvious the linear similarity between the images.

[0097] For image localization applications, the coordinates corresponding to the largest correlation coefficient in the correlation coefficient matrix are the localization coordinates of a small image or a pixel on a large image.

[0098] Although the correlation coefficient matching method has high accuracy, its computational complexity prevents its independent application in real-time image matching. Improvements and auxiliary constraints are needed to reduce computational load and meet the operational requirements of real-time image matching systems. This paper's main approach is to simplify the correlation coefficient formula without affecting its performance as a similarity measure, thereby achieving approximate calculations and improving computational speed and reliability.

[0099] According to statistical theory, the formula for calculating the correlation coefficient should be:

[0100]

[0101]

[0102]

[0103] The simplified formula is:

[0104]

[0105] Where f represents the image data of the previous image centered on the feature point, and g represents the image data of the corresponding search area in the later image. The mean and sum of g during a single matching process... It is a constant, which can be calculated first, and then simply substituted into the formula for calculation in subsequent matching to avoid duplication.

[0106] (ii) Phase Consistency Method:

[0107] Phase consistency-based heterogeneous image matching technology comprehensively utilizes the intensity and orientation information of phase consistency to construct a descriptor representing the geometric structural features of the image. The histogram of oriented phase congruency (HOPC) feature descriptor can obtain the structural attributes of the image. In addition, a similarity measure called HOPCncc is used, and the orthogonalized correlation coefficient (NCC) of the HOPC descriptor is used for multi-mode registration.

[0108] HOPC divides the template window into several blocks, calculates the phase consistency orientation histogram of each block, and links them together to form the final feature description vector. See also Figure 3 The diagram shows a flowchart of a HOPC algorithm. The HOPC extraction process mainly includes the following steps:

[0109] ① Select a template window of a certain size on the image.

[0110] ② Within the template window, calculate the phase consistency intensity value and direction of each pixel to provide feature information for the construction of HOPC.

[0111] ③ Divide the template window into several blocks, each block containing several cells, forming the basic structure of HOPC.

[0112] ④ Calculate the phase consistency direction histogram of the block and cell, and perform normalization to eliminate the influence of illumination changes.

[0113] ⑤ Collect the gradient direction histogram vectors within all blocks together to form the HOPC feature vector describing the entire template window.

[0114] HOPC is a feature descriptor for calculating the internal structure of an image. Because structural properties are more independent of the image's grayscale distribution, this descriptor can be used to match two images with similar shapes but significant nonlinear radiometric differences. Therefore, the NCC of the HOPC descriptor can be used as a similarity measure (called HOPCncc) for image matching, defined as:

[0115]

[0116] In the formula, VA and VB are the HOPC descriptors of image region A and image region B, respectively. and These represent the average values ​​of VA and VB, respectively.

[0117] During template matching, a template window moves pixel by pixel within an image or a search area, calculating the HOPCncc for each template window to be matched. Since most pixels of adjacent template windows overlap, this requires many repetitive calculations. This problem can be solved by utilizing the fast matching mechanism of HOPCncc.

[0118] Homogeneous point detection using HOPCncc involves two steps: extracting HOPC descriptors and calculating the NCC between descriptors. The first step consumes a significant amount of time in the matching process. To extract HOPC descriptors, the template window is divided into overlapping blocks, and the descriptors of these blocks are calculated to form the final dense descriptor. Therefore, a block can serve as the basic element of a HOPC descriptor. To reduce the computation time of template matching, a block region centered at each pixel in the image needs to be defined, and the HOPC descriptor of each block is extracted. Each pixel will have a block's HOPC descriptor, thus forming a three-dimensional descriptor for the entire image, called a block-HOPC image. Then, the block-HOPC descriptors are collected at intervals of several pixels to obtain the HOPC descriptor of the template window.

[0119] To ensure the accuracy of the initial matching points, each feature point to be matched must be successfully matched by both the correlation coefficient method and the HOPC phase correlation matching method before it can be identified as an initial matching point. The final matching point format is (x1, y1, x2, y2), where x1 and y1 represent the row and column numbers of the previous image, and x2 and y2 represent the row and column numbers of the later image.

[0120] Step 4: For each feature point to be matched, if the matching result of the feature point to be matched is consistent in the matching point set corresponding to each traversal process, then retain the matching feature point and the matching result of the matching feature point to obtain the initial matching point set.

[0121] For example, suppose that the matching result of a certain feature point to be matched is consistent in the matching point set after traversing the sliding window of three sizes (large, medium, and small), indicating that the matching result of the feature point to be matched is highly accurate. At this time, the feature point to be matched and its matching result can be saved as the initial matching point set.

[0122] This invention proposes the concept of initial matching points. By utilizing two existing matching methods simultaneously, it ensures the accuracy of initial matching points and reduces the probability of incorrect points. Since the initial accuracy of images in different stages may not be particularly good, a pyramid-based step-by-step matching system combined with a large search radius is used to ensure the stability of the initial matching points.

[0123] Regarding step S106 above, this embodiment of the invention provides an implementation method for constructing an irregular triangular mesh based on an initial set of matching points. This can be accomplished using the CGAL library. Experiments have shown that the CGAL library outperforms other libraries in both mesh construction efficiency and quantity. The irregular triangular mesh can be an irregular Delaunay triangular mesh, such as... Figure 4 The diagram shows an irregular Delaunay triangulation.

[0124] Regarding the aforementioned step S106, this embodiment of the invention also provides an implementation method for densely matching the pixels inside each triangle in the irregular triangular network to obtain a set of target densely matched points between the earlier and later remote sensing images, as shown in steps a to c below:

[0125] Step a: Perform dense matching on the pixels inside each triangle in the irregular triangular network to obtain the initial dense matching point set between the previous and later remote sensing images.

[0126] Each triangulation consists of three initial collocation points, and these three points are just enough to solve for the affine transformation parameters. The function of affine transformation is a linear transformation between one two-dimensional coordinate (early remote sensing image) and another two-dimensional coordinate (later remote sensing image). It uses the initial collocation point data to calculate the affine transformation coefficients, i.e., 6 parameters, and requires at least three initial collocation points to solve.

[0127] The formula for the affine transformation model is as follows:

[0128] f(x) = a0x + a1y + a2

[0129] f(y) = b0x + b1y + b2

[0130] The above equation describes the affine transformation relationship from two-dimensional points x and y to f(x) and f(y). The six parameters of the affine transformation, a0, a1, a2, b0, b1, and b2, are the objects to be solved by the system of equations.

[0131] Since the initial point matching has established the coordinate affine transformation relationship between the earlier and later images, the dense matching method within the triangulation differs from the initial point matching method:

[0132] (1) The matching method can use only one of the matching algorithms, either correlation coefficient or phase consistency, which can save matching time;

[0133] (2) The most time-consuming factor - the search radius R can be adjusted to 30 or even smaller. With the accuracy guarantee of the three initial matching points of the triangulation, there is no need to consider the initial accuracy difference between the early and later images.

[0134] (3) The density of matching can be set to a parameter P. Matching every pixel is 1, and matching every other pixel is 2. The denser the matching, the more accurate the final determination of the change area, but the longer the matching time. A value can be determined based on the requirements and calculation time.

[0135] Taking pixel-by-pixel dense matching as an example, for each pixel (x, y) within the triangulation, the affine transformation parameters calculated using the corresponding triangle can be used to calculate the initial pixel coordinates (f(x), f(y)) of the corresponding later image. Data f from the earlier image, centered on the pixel (x, y) with an outer radius of R1, and data g from the later image, centered on the corresponding pixel (f(x), f(y)) with an outer radius of R2, are read. R1 must be less than R, allowing f to slide along g to calculate the correlation coefficient. Finally, the position coordinates of f in g, where the correlation coefficient is at its maximum, are calculated, yielding the matching point of that pixel. The affine transformation parameters of the triangulation provide a relatively accurate range of initial matching coordinates, while the correlation coefficient matching solves for the precise coordinates.

[0136] In this embodiment of the invention, an irregular triangular network is established using initial collocation points. Three points can be used to determine a coordinate transformation relationship from one surface to another (i.e., an affine transformation model). Then, dense matching is performed on the pixels inside each triangle. Since there is an affine transformation model constructed with initial collocation points, the search position of the subsequent remote sensing image can be located relatively accurately through the affine transformation model. Therefore, the matching inside the triangle can use a single correlation coefficient method, and there is no need to use a pyramid. The search radius can be very small, which can greatly speed up the matching efficiency.

[0137] Without the construction of initial matching points and triangulation meshes, the time required to complete dense matching of all pixels in two images with large initial accuracy differences between the pre- and post-processing images would be enormous, requiring a very large search radius. Therefore, the construction of the initial matching point set and triangulation meshes is the core content of the embodiments of this invention.

[0138] Step b: Determine the affine transformation error value of each densely matched point in the initial densely matched point set. If the affine transformation error value of the densely matched point is greater than the preset error threshold, then remove the densely matched point from the initial densely matched point set to obtain the intermediate densely matched point set.

[0139] In one implementation, after the dense matching of the triangulation network is completed, a very dense set of points (x1, y1, x2, y2, ...) will be formed. At this time, the error value of each densely matched point can be calculated according to the affine parameters formed by the initial matching points (including a0, a1, a2, b0, b1, b2), and the incorrect matching points can be deleted according to the error threshold.

[0140] In practical implementation, the affine transformation error value of all dense points within a triangle can be calculated on a unit basis. The affine transformation error value is obtained by substituting the densely matched points into the formula of the affine transformation model to obtain the corresponding f(x) and f(y) values. The difference between the obtained f(x) and f(y) values ​​and the true f(x) and f(y) values ​​is then calculated to obtain the affine transformation error value of that densely matched point. Further, the affine transformation error value of this densely matched point is compared with a preset error threshold. If it exceeds the preset error threshold, the densely matched point is removed from the initial set of densely matched points; otherwise, it is retained, thus obtaining an intermediate set of densely matched points.

[0141] Step c: Detect missing matching points in the intermediate dense matching point set, and perform nearest neighbor matching on the missing matching points to obtain the target dense matching point set between the early and later remote sensing images.

[0142] In practical applications, dense matching constrained by triangulation obtains matching points for most pixels. However, since the affine parameters calculated by the triangulation may not fully represent the transformation of the entire triangular region, and some matching points were deleted in the previous step, and some areas that have indeed changed cannot be matched successfully (these are the areas of change we ultimately need), some pixels will not be matched successfully and require further matching processing. The purpose of this step is to extract as many areas as possible from the previous and later remote sensing images that could not be matched successfully.

[0143] Nearest neighbor matching for missing match points mainly consists of two steps:

[0144] (a) Detect missing matches in the intermediate dense matching point set, including (a1) to (a2):

[0145] (a1) Divide the set of densely matched points in the middle into a set of densely matched points corresponding to each triangle in the irregular triangular network;

[0146] (a2) For each triangle in the cyclic irregular triangular mesh, determine the unmatched pixels in the dense matching point subset corresponding to each triangle according to the preset dense matching degree, and determine the unmatched pixels as missing matching points.

[0147] In one implementation, each triangle can be looped through, and based on the set density matching degree P, it can be determined which row and column numbers of the points have not been matched successfully, and the row and column numbers of the pixels can be recorded.

[0148] For example, if the density of the match P is set to 1, then every pixel needs to be matched successfully, and any unmatched pixels should be recorded. If the density of the match P is set to 2, then every other pixel needs to be matched successfully. If the density of the match P is set to x, then every x-1 pixels needs to be matched successfully, otherwise the row and column numbers of the unmatched pixels are recorded.

[0149] (b) Nearest neighbor matching is performed on the missing matching points to obtain a set of densely matched target points between the earlier and later remote sensing images, including (b1) to (b3):

[0150] (b1) Determine the distance between the missing matching point and each successfully matched pixel, and determine multiple target pixels from each successfully matched pixel based on the distance.

[0151] In one implementation, several pixels closest to the missed matching points can be selected from the successfully matched pixels as target pixels. The number of target pixels is at least three.

[0152] (b2) Construct an affine transformation model based on each target pixel.

[0153] (b3) Feature point matching is performed on the missing matching points by using an affine transformation model to obtain a set of densely matched target points between the early and later remote sensing images.

[0154] In one implementation, an affine transformation model can be used to automatically rematch the missing matching point to obtain the matching result of the missing matching point. The matching result of the missing matching point can be saved to the intermediate dense matching point set to obtain the target dense matching point set.

[0155] This embodiment of the invention performs a step of detecting missing matching points and performing nearest neighbor matching on the missing matching points, so as to match as many pixels as possible that can be matched, thereby enabling more accurate detection of changing regions.

[0156] Regarding the aforementioned step S108, this embodiment of the invention provides an implementation method for determining the change region between earlier and later remote sensing images based on a set of densely matched target points, as shown in steps one through four below:

[0157] Step 1: Divide the preliminary remote sensing images into multiple image blocks.

[0158] In one implementation, (1) the previous image is divided into blocks. The block size BLOCKSIZE determines the granularity of the extracted change area. The smaller the value, the finer the granularity. It can be set according to the requirements, usually 512.

[0159] Step 2: Based on the target dense matching point set, determine the number of dense matching points contained in each image block.

[0160] Step 3: Determine the matching success rate of each image block based on the total number of pixels, the number of invalid pixels, the number of dense matching points, and the preset density matching degree of the previous remote sensing images.

[0161] In one implementation, the difference between the total number of pixels N and the number of invalid pixels M in the previous remote sensing image is determined; for each image block, the product between the number of densely matched points C contained in the image block and the preset dense matching degree P is determined; the ratio of the difference to the product is determined as the matching success rate of the image block.

[0162] Specifically, invalid values ​​are removed from each image block, and the matching success rate of each image block is calculated. For example, if the image block is divided into 512 x 512 blocks, the total number of pixels is N (262,144), invalid values ​​occupy M pixels, the number of successfully matched points is C, and the density of matching is P, then the matching success rate F of each image block is:

[0163]

[0164] Where F is a floating-point number ranging from 0 to 1.

[0165] Step 4: If the matching success rate of the image patch is less than the preset success rate threshold, then the image patch is determined to be the change area between the previous remote sensing image and the later remote sensing image.

[0166] In one implementation, a change threshold Q is set. When F is greater than Q, it means that the image block has no change. When F is less than Q, it means that the image block is a changed area. All changed areas are output as vectors and saved, which is the final result of the changed areas.

[0167] This invention utilizes the characteristic that regions without change can be successfully matched while regions with change cannot be successfully matched to solve the problem of determining the change regions in unregistered pre- and post-registration remote sensing images using automatic matching technology, which is also the core content of this invention.

[0168] In summary, the remote sensing image change region determination method based on triangular mesh dense matching provided in this embodiment of the invention has at least the following characteristics:

[0169] (1) Using the characteristic that areas without change can be successfully matched while areas with change cannot be successfully matched, automatic matching technology is used to solve the problem of determining the change areas in unregistered remote sensing images before and after registration.

[0170] (2) The concept of initial matching points is proposed, which ensures the accuracy of initial matching points and reduces the probability of mismatches by utilizing the two existing matching methods simultaneously. Since the initial accuracy of images in the early and late stages may not be particularly good, the stability of initial matching points is ensured by using pyramid-level matching and a large search radius;

[0171] (3) Use the initial matching points to establish an irregular triangular network. The coordinate transformation relationship from one face to another can be determined by the three points (affine transformation model). Then perform dense matching on the pixels inside each triangle. Since there is an affine model constructed by the initial matching points, the search position of the later image can be located more accurately through the affine model. Therefore, the matching inside the triangle can use a single correlation coefficient method and does not need to use a pyramid. The search radius can be very small, which can greatly speed up the matching efficiency.

[0172] (4) In addition, missing matching points are detected and nearest neighbor matching is performed on the missing matching points to match as many pixels as possible that can be matched, so that the changing areas can be detected more accurately.

[0173] Remote sensing image data, especially satellite remote sensing image data, is extremely large, with a single image covering an area of ​​at least tens or even hundreds of square kilometers. The demand for change detection in applications such as land use is consistently high. While manual identification of change areas is accurate, it is inadequate for the sheer volume of remote sensing data. The method provided in this invention can largely free up human hands and eyes, quickly locating the rectangular areas of change in remote sensing images. These areas can then be manually interpreted or processed by other algorithms capable of accurately extracting the boundaries of the changes.

[0174] For ease of understanding, this invention provides another method for determining change regions in remote sensing images based on dense triangular mesh matching, see [link to relevant documentation]. Figure 5 The flowchart shown is for another method for determining the change area of ​​remote sensing images based on dense triangular mesh matching. This method mainly includes the following steps S502 to S510:

[0175] Step S502, initial feature point matching;

[0176] Step S504: Initial point triangulation construction;

[0177] Step S506, dense matching of triangular mesh constraints;

[0178] Step S508, nearest neighbor matching for missing matching points;

[0179] Step S510: Determine the changing rectangular region.

[0180] In summary, the remote sensing image change region determination method based on triangular mesh dense matching provided in this invention utilizes the characteristic that unchanged regions can be successfully matched while changed regions cannot. By employing matching strategies such as triangular mesh constraints, it can automatically and quickly extract the changed rectangular regions from previous and later remote sensing images. Although it cannot accurately extract the boundaries of the changes, it is still of great significance as an algorithm for initially determining the range of changes. It can save a lot of manual labor costs in searching for changed regions in massive amounts of previous and later remote sensing images.

[0181] Regarding the remote sensing image change region determination method based on triangulation dense matching provided in the foregoing embodiments, this invention provides a remote sensing image change region determination device based on triangulation dense matching, see [link to relevant documentation]. Figure 6 The diagram shows a structural schematic of a remote sensing image change region determination device based on triangular mesh dense matching. The device mainly includes the following parts:

[0182] The image acquisition module 602 is used to acquire the pre-processed remote sensing images and post-processed remote sensing images to be processed;

[0183] The initial matching module 604 is used to perform feature point matching between the early and later remote sensing images to obtain an initial set of matching points between the early and later remote sensing images.

[0184] The dense matching module 606 is used to construct an irregular triangular network based on the initial matching point set, and to perform dense matching on the pixels inside each triangle in the irregular triangular network to obtain the target dense matching point set between the previous remote sensing image and the later remote sensing image.

[0185] The change area determination module 608 is used to determine the change area between previous and later remote sensing images based on the set of densely matched target points.

[0186] The remote sensing image change region determination device based on dense triangulation matching provided in this invention utilizes the characteristic that unchanged regions can be successfully matched while changed regions cannot. This device can quickly and stably determine the range of changed regions in remote sensing images, enabling refined automatic change detection or manual judgment. This invention is applicable not only to multi-band images but also to single-band images. Furthermore, this invention does not require high-precision registration of pre- and post-concurrent remote sensing images, nor does it require human intervention or training with a large number of samples. It exhibits high stability and robustness and can automatically extract changed regions from multi-temporal remote sensing images in batches.

[0187] In one implementation, the initial matching module 604 is further configured to:

[0188] Construct an early pyramid image corresponding to the early remote sensing image and a later pyramid image corresponding to the later remote sensing image; both the early and later pyramid images include multiple image layers.

[0189] Multiple feature points to be matched are extracted from previous remote sensing images;

[0190] By using sliding windows of different sizes, each image layer of the early pyramid image and each image layer of the later pyramid image are traversed multiple times. In each traversal, multiple feature point matching algorithms are used to perform feature point matching for each feature point to be matched, so as to obtain the matching point set corresponding to each traversal.

[0191] For each feature point to be matched, if the matching result of the feature point to be matched is consistent in the matching point set corresponding to each traversal process, then the matching feature point and the matching result of the matching feature point are retained to obtain the initial matching point set.

[0192] In one implementation, the dense matching module 606 is further configured to:

[0193] Dense matching is performed on the pixels inside each triangle in the irregular triangular network to obtain the initial dense matching point set between the previous and later remote sensing images.

[0194] Determine the affine transformation error value of each densely matched point in the initial densely matched point set. If the affine transformation error value of a densely matched point is greater than a preset error threshold, then remove the densely matched point from the initial densely matched point set to obtain an intermediate densely matched point set.

[0195] Detect missing matching points in the intermediate dense matching point set, and perform nearest neighbor matching on the missing matching points to obtain the target dense matching point set between the earlier and later remote sensing images.

[0196] In one implementation, the dense matching module 606 is further configured to:

[0197] The set of densely matched points in the middle is divided into a set of densely matched points corresponding to each triangle in the irregular triangular network;

[0198] For each triangle in the cyclic irregular triangular mesh, the pixels that were not successfully matched in the dense matching point subset corresponding to each triangle are determined according to the preset dense matching degree, and the pixels that were not successfully matched are determined as missing matching points.

[0199] In one implementation, the dense matching module 606 is further configured to:

[0200] Determine the distance between the missed matching point and each successfully matched pixel, and determine multiple target pixels from each successfully matched pixel based on the distance;

[0201] Construct an affine transformation model based on each target pixel;

[0202] By using an affine transformation model to perform feature point matching on the missing matching points, a set of densely matched target points between the earlier and later remote sensing images can be obtained.

[0203] In one implementation, the change region determination module 608 is further configured to:

[0204] The previous remote sensing images were divided into multiple image blocks;

[0205] Based on the target dense matching point set, determine the number of dense matching points contained in each image block;

[0206] The matching success rate of each image block is determined based on the total number of pixels, the number of invalid pixels, the number of dense matching points, and the preset density matching degree of the previous remote sensing images.

[0207] If the matching success rate of an image patch is less than a preset success rate threshold, then the image patch is determined to be the area of ​​change between earlier and later remote sensing images.

[0208] In one implementation, the change region determination module 608 is further configured to:

[0209] Determine the difference between the total number of pixels and the number of invalid pixels in the previous remote sensing image;

[0210] For each image block, determine the product between the number of densely matched points contained in the image block and the preset dense matching degree;

[0211] The ratio of the difference to the product is determined as the matching success rate of the image block.

[0212] The device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0213] This invention provides an electronic device, specifically, the electronic device includes a processor and a storage device; the storage device stores a computer program, and the computer program, when run by the processor, executes the method described in any of the above embodiments.

[0214] Figure 7The present invention provides a schematic diagram of the structure of an electronic device 100, which includes a processor 70, a memory 71, a bus 72 and a communication interface 73. The processor 70, the communication interface 73 and the memory 71 are connected through the bus 72. The processor 70 is used to execute executable modules, such as computer programs, stored in the memory 71.

[0215] The memory 71 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 73 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0216] Bus 72 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0217] The memory 71 is used to store programs. After receiving an execution instruction, the processor 70 executes the programs. The method executed by the device for defining the flow process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 70 or implemented by the processor 70.

[0218] The processor 70 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 70 or by instructions in software form. The processor 70 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 71. Processor 70 reads the information in memory 71 and, in conjunction with its hardware, completes the steps of the above method.

[0219] The computer program product of the readable storage medium provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For specific implementation, please refer to the foregoing method embodiments, which will not be repeated here.

[0220] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0221] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for determining change regions in remote sensing images based on dense triangular mesh matching, characterized in that, include: Acquire the pre-processed and post-processed remote sensing images to be processed; Feature point matching is performed on the earlier remote sensing images and the later remote sensing images to obtain an initial set of matching points between the earlier and later remote sensing images; An irregular triangular network is constructed based on the initial set of matching points, and dense matching is performed on the pixels inside each triangle in the irregular triangular network to obtain the target dense matching point set between the earlier remote sensing image and the later remote sensing image. Based on the target dense matching point set, determining the change region between the previous remote sensing image and the later remote sensing image includes: dividing the previous remote sensing image into multiple image blocks; determining the number of dense matching points contained in each image block based on the target dense matching point set; determining the matching success rate of each image block according to the total number of pixels, the number of invalid pixels, the number of dense matching points, and a preset dense matching degree; if the matching success rate of the image block is less than a preset success rate threshold, then the image block is determined to be the change region between the previous remote sensing image and the later remote sensing image.

2. The method for determining the change region of remote sensing images based on dense triangular mesh matching according to claim 1, characterized in that, Feature point matching is performed on the earlier and later remote sensing images to obtain an initial set of matching points between the earlier and later remote sensing images, including: Construct an early pyramid image corresponding to the early remote sensing image and a later pyramid image corresponding to the later remote sensing image; wherein, both the early pyramid image and the later pyramid image include multiple image layers; Multiple feature points to be matched are extracted from the aforementioned previous remote sensing images; By using sliding windows of different sizes, each image layer of the early pyramid image and each image layer of the later pyramid image are traversed multiple times, so that each feature point to be matched is matched using multiple feature point matching algorithms in each traversal process, and the matching point set corresponding to each traversal process is obtained. For each feature point to be matched, if the matching result of the feature point to be matched is consistent in the matching point set corresponding to each traversal process, then the matching feature point and the matching result of the matching feature point are retained to obtain the initial matching point set.

3. The method for determining the change region of remote sensing images based on dense triangular mesh matching according to claim 1, characterized in that, Dense matching is performed on the pixels inside each triangle in the irregular triangular network to obtain a set of target dense matching points between the earlier and later remote sensing images, including: Dense matching is performed on the pixels inside each triangle in the irregular triangular network to obtain an initial set of densely matched points between the earlier remote sensing image and the later remote sensing image. Determine the affine transformation error value of each densely matched point in the initial densely matched point set. If the affine transformation error value of the densely matched point is greater than a preset error threshold, then remove the densely matched point from the initial densely matched point set to obtain an intermediate densely matched point set. Detect missing matching points in the intermediate dense matching point set, and perform nearest neighbor matching on the missing matching points to obtain the target dense matching point set between the earlier remote sensing image and the later remote sensing image.

4. The method for determining the change region of remote sensing images based on dense triangular mesh matching according to claim 3, characterized in that, Detecting missing matching points in the intermediate dense matching point set includes: The intermediate dense matching point set is divided into a dense matching point subset corresponding to each triangle in the irregular triangular network; The process involves iterating through each triangle in the irregular triangular mesh, determining the unmatched pixels in the dense matching point subset corresponding to each triangle based on a preset density matching degree, and identifying the unmatched pixels as missing matching points.

5. The method for determining the change region of remote sensing images based on dense triangular mesh matching according to claim 3, characterized in that, Nearest neighbor matching is performed on the missing matching points to obtain a set of densely matched target points between the earlier and later remote sensing images, including: Determine the distance between the missed matching point and each successfully matched pixel, and determine multiple target pixels from each successfully matched pixel based on the distance; Construct an affine transformation model based on each of the target pixels; The missing matching points are matched using the affine transformation model to obtain a set of densely matched target points between the earlier and later remote sensing images.

6. The method for determining the change region of remote sensing images based on dense triangular mesh matching according to claim 1, characterized in that, Based on the total number of pixels, the number of invalid pixels, the number of densely matched points, and the preset density of matching in the previous remote sensing images, the matching success rate of each image block is determined, including: Determine the difference between the total number of pixels and the number of invalid pixels in the previous remote sensing image; For each image block, determine the product between the number of dense matching points contained in the image block and a preset dense matching degree; The ratio of the difference to the product is determined as the matching success rate of the image block.

7. A device for determining the change region of remote sensing images based on dense triangular mesh matching, characterized in that, include: The image acquisition module is used to acquire pre-processed and post-processed remote sensing images to be processed. The initial matching module is used to perform feature point matching on the earlier remote sensing images and the later remote sensing images to obtain an initial set of matching points between the earlier remote sensing images and the later remote sensing images. The dense matching module is used to construct an irregular triangular network based on the initial matching point set, and to perform dense matching on the pixels inside each triangle in the irregular triangular network to obtain the target dense matching point set between the previous remote sensing image and the later remote sensing image. The change region determination module is used to determine the change region between the previous remote sensing image and the later remote sensing image based on the target dense matching point set, including: dividing the previous remote sensing image into multiple image blocks; determining the number of dense matching points contained in each image block based on the target dense matching point set; determining the matching success rate of each image block according to the total number of pixels, the number of invalid pixels, the number of dense matching points, and a preset dense matching degree of the previous remote sensing image; if the matching success rate of the image block is less than a preset success rate threshold, then the image block is determined to be the change region between the previous remote sensing image and the later remote sensing image.

8. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 6.

Images