A remote sensing image matching method for complex environments

Abstract

The present application relates to the technical field of remote sensing image matching, and specifically to a remote sensing image matching method for complex environments, the method comprising: obtaining multiple remote sensing images of the same target area; selecting multiple feature points in each remote sensing image; recording each pixel point in the local window of each feature point as a candidate pixel point; determining the grayscale deviation of each candidate pixel point; obtaining each significant pixel point in the local neighborhood of each candidate pixel point; analyzing the distance relationship between all the significant pixel points of each candidate pixel point and the fitting straight line, and determining the smoothing weight of each candidate pixel point; obtaining the structural prominence of each candidate pixel point; determining each structural attribute pixel point of the local window of each feature point; determining each point pair of each feature point based on the structural attribute pixel point, obtaining the feature descriptor of each feature point, and obtaining the matching results of multiple remote sensing images using an image matching algorithm. The present application improves the accuracy of remote sensing image matching.

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing image matching, and in particular to a remote sensing image matching method for complex environments. Background Art

[0002] Remote sensing images refer to image data in a specific scene obtained through sensor technology. Remote sensing image matching refers to mapping an image to one or more images in the same scene to establish a mapping relationship between two or more remote sensing images.

[0003] The ORB algorithm (Oriented FAST and Rotated BRIEF) is a visual feature extraction and matching algorithm that is highly efficient, rotationally invariant, and scale invariant, and is often used for remote sensing image matching. However, when the ORB algorithm constructs feature descriptors for feature points, each pair of feature points is selected through random sampling, resulting in insufficient distinguishing ability of the feature descriptors for feature points and low image matching accuracy. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides a remote sensing image matching method for complex environments to solve the existing problems.

[0005] The remote sensing image matching method for complex environments in this application adopts the following technical solutions:

[0006] An embodiment of the present application provides a remote sensing image matching method for complex environments, the method comprising the following steps:

[0007] Acquire multiple remote sensing images of the same target area; select multiple feature points in each remote sensing image;

[0008] Each pixel point in the local window of each feature point is recorded as a candidate pixel point; the grayscale difference of the pixel points in the local neighborhood of each candidate pixel point is analyzed to determine the grayscale deviation of each candidate pixel point; based on the magnitude relationship between the grayscale difference of each pixel point in the local neighborhood of each candidate pixel point and the adjacent pixel points, each significant pixel point in the local neighborhood of each candidate pixel point is obtained;

[0009] Based on the position coordinates of all significant pixels on the remote sensing image, a fitting straight line is obtained by fitting; the distance relationship between all the significant pixels of each candidate pixel and the fitting straight line is analyzed to determine the smoothing weight of each candidate pixel;

[0010] Combining the grayscale deviation with the smoothing weight, the structural prominence of each candidate pixel is obtained; for each feature point, the candidate pixel whose structural prominence in the local window is greater than or equal to a preset percentage of the maximum value of all the structural prominences in the local window is recorded as a potential structural pixel, and the potential structural pixel points in the local window are clustered to obtain the real structural attribute cluster and each cluster to be screened;

[0011] The differences between the grayscale values ​​of the pixels in each cluster to be screened and the real structural attribute cluster, as well as the distance relationship between the pixels in each cluster to be screened and the real structural attribute cluster are analyzed to determine the spatial grayscale deviation of each pixel in each cluster to be screened, and the structural attribute pixel points of the local window of each feature point are obtained in combination with the spatial grayscale deviation and the real structural attribute cluster; based on the structural attribute pixel points, each point pair of each feature point is determined, each point pair is marked to obtain a feature descriptor of each feature point, and the matching results of the multiple remote sensing images are obtained using an image matching algorithm.

[0012] In one embodiment, the grayscale histogram of all pixels in the local neighborhood of each candidate pixel is fitted to obtain a fitting curve, and the grayscale deviation is the difference between the grayscale level of the maximum maximum value and the grayscale level of the second maximum maximum value on the fitting curve.

[0013] In one embodiment, the process of determining each significant pixel point is as follows:

[0014] The mean of the grayscale difference between each pixel and the adjacent pixel in the local neighborhood of each candidate pixel is calculated, the mean values ​​of all the pixels in the local neighborhood of each candidate pixel are arranged in descending order, and a preset number of pixels after the arrangement are used as the significant pixels.

[0015] In one embodiment, the smoothing weight is determined as follows:

[0016] The mean of the metric distances between all the significant pixels of each candidate pixel and the fitting straight line is calculated and recorded as the first mean, and the smoothing weight is the reciprocal of the sum of the first mean and a preset value greater than 0.

[0017] In one embodiment, the structural prominence is the product of the grayscale deviation and the smoothing weight.

[0018] In one embodiment, the process of determining the real structural attribute cluster and each cluster to be screened is as follows:

[0019] The real structural attribute cluster is the cluster containing the most potential structural pixels, and the clusters to be screened are the remaining clusters except the real structural attribute cluster.

[0020] In one embodiment, the process of determining the spatial grayscale deviation is:

[0021] For the local window of each feature point, the mean of the grayscale values ​​of all pixels in the real structural attribute cluster is calculated, which is recorded as the second mean. The difference between the grayscale value of each pixel in each cluster to be screened and the second mean is calculated, which is recorded as the selected difference. The metric distance between each pixel in each cluster to be screened and the cluster center of the real structural attribute cluster is calculated, which is recorded as the selected distance. The spatial grayscale deviation of each pixel in each cluster to be screened is the product of the selected difference and the selected distance.

[0022] In one embodiment, the process of determining each structural attribute pixel point is as follows:

[0023] The spatial grayscale deviation of all pixels in the clusters to be screened is subjected to threshold segmentation, and the pixels in the clusters to be screened whose spatial grayscale deviation is less than the segmentation threshold and all the pixels in the real structural attribute cluster are determined as structural attribute pixels.

[0024] In one embodiment, determining each point pair of each feature point based on the structural attribute pixel point includes:

[0025] The pixels other than the structural attribute pixels in the local window of each feature point are recorded as background pixels. Any structural attribute pixel and any background pixel in the local window of each feature point form a point pair.

[0026] In one embodiment, the process of determining the feature descriptor of each feature point is as follows:

[0027] If the grayscale value of the structural attribute pixel in each point pair is greater than the grayscale value of the background pixel, the point pair is marked as 1, otherwise, it is marked as 0. The marking values ​​of all point pairs of each feature point form the feature descriptor of each feature point.

[0028] This application has at least the following beneficial effects:

[0029] The present application obtains multiple remote sensing images of the same target area; selects multiple feature points in each remote sensing image; records each pixel point in the local window of each feature point as a candidate pixel point; analyzes the grayscale difference of the pixel points in the local neighborhood of each candidate pixel point, and determines the grayscale deviation of each candidate pixel point; the grayscale deviation reflects the polarization degree of the grayscale of the pixel points in the local neighborhood of the candidate pixel point, embodies the possibility that the candidate pixel point is a structural attribute pixel point, and improves the reliability of the determination of the structural attribute pixel point; based on the size relationship between the grayscale difference of each pixel point in the local neighborhood of each candidate pixel point and the adjacent pixel points, obtains each significant pixel point in the local neighborhood of each candidate pixel point; fits the position coordinates of all significant pixels on the remote sensing image to obtain a fitting line; analyzes the distance relationship between all the significant pixels of each candidate pixel point and the fitting line, Determine the smoothing weight of each candidate pixel; the smoothing weight reflects the linearity of the distribution of significant pixels in the local neighborhood of the candidate pixel, combined with the distribution characteristics of the structural attribute pixels, and also reflects the possibility that the candidate pixel is a structural attribute pixel; combine the grayscale deviation and the smoothing weight to obtain the structural prominence of each candidate pixel; filter the structural attribute pixels in the candidate pixels according to the structural prominence, thereby improving the accuracy of determining the structural attribute pixels; obtain the structural attribute pixels of the local window of each feature point based on the structural prominence; determine the point pairs of each feature point based on the structural attribute pixels, obtain the feature descriptor of each feature point, improve the accuracy of the descriptor in representing the feature point, and use the image matching algorithm to obtain the matching results of the multiple remote sensing images, improve the matching accuracy of the remote sensing images, and reduce the probability of mismatching. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 A flowchart of the steps of a remote sensing image matching method for complex environments provided in this application;

[0032] Figure 2 Get the flow chart for the structural attribute pixels;

[0033] Figure 3 is the matching result of the optical remote sensing image and the SAR remote sensing image in this embodiment;

[0034] Figure 4This is the matching result of the optical remote sensing image and the SAR remote sensing image obtained using the traditional ORB algorithm. DETAILED DESCRIPTION

[0035] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following is a detailed description of the remote sensing image matching method for complex environments proposed in the present application, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0036] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0037] The following is a detailed description of a remote sensing image matching method for complex environments provided by the present application in conjunction with the accompanying drawings.

[0038] An embodiment of the present application provides a remote sensing image matching method for complex environments. Specifically, the following remote sensing image matching method for complex environments is provided. Figure 1 , the method comprises the following steps:

[0039] Step S001, obtaining multiple remote sensing images of the same target area; selecting multiple feature points in each remote sensing image.

[0040] Since optical remote sensing images and SAR (Synthetic Aperture Radar) remote sensing images account for a large proportion of remote sensing images, this embodiment uses optical remote sensing and SAR remote sensing images as remote sensing images to be matched. This embodiment takes urban area A with a complex environment as the target area, uses drones mounted with optical high-definition sensors and SAR, and collects optical remote sensing images and SAR remote sensing images of the target area at the same altitude. In this embodiment, the altitude for collecting remote sensing images is 10KM, which can be set by the implementer according to the actual situation, and this embodiment does not limit it here.

[0041] Taking optical remote sensing images as an example, the Fast algorithm is mainly used to select feature points in the ORB algorithm. In this embodiment, the number of feature points selected is 5% of the image size, that is, , where H and W are the height and width of the optical remote sensing image, Round() is a rounding function, and N represents the number of feature points selected. Both the ORB algorithm and the Fast algorithm are existing well-known technologies, and will not be described in detail in this embodiment.

[0042] Step S002, record each pixel point in the local window of each feature point as a candidate pixel point; analyze the grayscale difference of the pixels in the local neighborhood of each candidate pixel point to determine the grayscale deviation of each candidate pixel point; based on the size relationship between the grayscale difference between each pixel point in the local neighborhood of each candidate pixel point and the adjacent pixel points, obtain each significant pixel point in the local neighborhood of each candidate pixel point.

[0043] When constructing feature descriptors for feature points in the traditional ORB algorithm, a single feature point is used as the center. Q groups of point pairs are randomly selected in the local window of , and binary descriptors are constructed based on the grayscale values ​​of the two points in each group of point pairs. Finally, descriptor matching is performed on remote sensing images of the same scene. The matching feature points are called homonymous points. Usually the default values ​​are S=31 and Q=256.

[0044] However, for heterogeneous remote sensing images, due to the different sensors used to collect data, nonlinear radiation differences occur, resulting in large variations in the grayscale differences of the same geographical location in different remote sensing images. Therefore, constructing feature descriptors based on the grayscale information of randomly selected point pairs will result in fewer matching points or a large number of mismatches for heterogeneous remote sensing images.

[0045] Although nonlinear radiation differences cause the grayscale values ​​of the same point in the target area to be different in different remote sensing images, since they are the same target area, even different remote sensing images have the same structural information, that is, the structural attributes of streets, houses, roads, etc. that appear in remote sensing images are the same. Therefore, the feature descriptors of feature points can be constructed by analyzing the same structural attributes in different remote sensing images.

[0046] Based on the above analysis, the feature points Each pixel point in the local window is recorded as a candidate pixel point. With each candidate pixel point as the center, a local neighborhood of each candidate pixel point is constructed. In this embodiment, the size of the local neighborhood is 7×7, which can be set by the implementer according to the actual situation. This embodiment does not limit it. All pixel points in the local neighborhood of each candidate pixel point are divided into gray levels, specifically: the gray values ​​0~255 of all pixels are quantized into 16 gray levels, the gray level corresponding to the pixel points with gray values ​​of 0~15 is 1, the gray level corresponding to the pixel points with gray values ​​of 16~31 is 2, and so on. The gray level of each pixel point in the local neighborhood of each candidate pixel point is obtained, the number of pixels in each gray level is counted, and the gray level histogram of all pixels in the local neighborhood of each candidate pixel point is constructed.

[0047] It should be noted that the number of gray levels can be set by the implementer according to the actual situation, and this embodiment does not limit it. The construction of the gray level histogram is an existing well-known technology, and this embodiment will not be described in detail.

[0048] The grayscale histogram of each candidate pixel is nonlinearly fitted using the least squares method to obtain the corresponding fitting curve, and the maximum values ​​on the fitting straight line are obtained. All the maximum values ​​are arranged from large to small, and the difference between the first maximum value and the second maximum value after arrangement is used as the grayscale deviation of each candidate pixel.

[0049] It should be understood that the difference represents the degree of difference between two variables, and can be calculated specifically by using the absolute value of the difference, the ratio, the square of the difference, etc. This embodiment uses the absolute value of the difference as the calculation method for the difference.

[0050] In this embodiment, the grayscale deviation of each candidate pixel can be calculated as follows:

[0051] ; Where B represents the grayscale deviation of the pixel to be selected, is the gray level corresponding to the first maximum value after the arrangement, is the gray level corresponding to the second maximum value, Indicates calculating the absolute value.

[0052] In this embodiment, the pixel points of structural attributes such as streets, houses, roads, etc. in the remote sensing image are recorded as structural attribute pixels. If the pixel point to be selected is a structural attribute pixel point, an obvious grayscale boundary will appear in the local neighborhood of the pixel point to be selected, corresponding to the background information and structural information in the remote sensing image, thereby causing two larger extreme value peaks to appear in the grayscale levels of all pixels in the local neighborhood of the pixel point to be selected. Since the grayscale difference between the background information and the structural information is large, the greater the difference between the grayscale levels corresponding to the first maximum value and the second maximum value, the greater the value of the grayscale deviation is.

[0053] Calculate the mean of the grayscale differences between each pixel in the local neighborhood of each candidate pixel and a preset number of adjacent pixels. In this embodiment, the mean of the absolute values ​​of the grayscale value differences between each pixel in the local neighborhood of each candidate pixel and its four adjacent pixels is calculated, and the mean values ​​of all the pixels in the local neighborhood of each candidate pixel are arranged from large to small, and the first 7 pixels after the arrangement are recorded as significant pixels.

[0054] It should be noted that the number of adjacent pixels and the number of significant pixels can be set by the implementer according to actual conditions, and this embodiment does not limit this.

[0055] Step S003, performing fitting based on the position coordinates of all significant pixels on the remote sensing image to obtain a fitting straight line; analyzing the distance relationship between all the significant pixels of each candidate pixel and the fitting straight line to determine the smoothing weight of each candidate pixel.

[0056] Based on the position coordinates of the significant pixels in the remote sensing image, all the significant pixels in the local neighborhood of each candidate pixel are linearly fitted using the least square method to obtain the corresponding fitting line, and the mean of the distances between all the significant pixels in the local neighborhood of each candidate pixel and the fitting line is calculated and recorded as the first mean. The distance between the significant pixel and the fitting line is the distance between the point and the line, and the calculation method is the existing known technology.

[0057] The smoothing weight of each candidate pixel is constructed based on the first mean value, and the specific calculation method is:

[0058] ; C is the smoothing weight of the pixel to be selected, is the first mean value of the pixel to be selected, To preset a value greater than 0 to avoid the denominator being 0, in this embodiment The implementer can set it according to the actual situation, and this embodiment does not limit it here.

[0059] If the pixel to be selected is a structural attribute pixel, the grayscale difference between the pixel on the boundary and the adjacent pixel in its local neighborhood is large, and it is easier to be screened out to obtain a significant pixel. The coordinates of the significant pixel are linearly fitted, and the pixel points on the boundary basically fall on the fitting line, so that the smaller the first mean obtained, the larger the value of the smoothing weight is obtained.

[0060] Step S004, combining the grayscale deviation and the smoothing weight to obtain the structural prominence of each candidate pixel point; based on the structural prominence, screening to obtain each structural attribute pixel point of the local window of each feature point; based on the structural attribute pixel point, determining each point pair of each feature point, obtaining the feature descriptor of each feature point, and using the image matching algorithm to obtain the matching results of the multiple remote sensing images.

[0061] According to the grayscale deviation and smoothing weight of each candidate pixel, the structural prominence A of each candidate pixel is obtained, which is specifically:

[0062] ; Where A represents the structural prominence of the candidate pixel, B represents the grayscale deviation of the candidate pixel, and C is the smoothing weight of the candidate pixel.

[0063] If the pixel to be selected is not a structural attribute pixel, the pixels in its local neighborhood mainly contain background information. Since the scene content corresponding to the background information may be relatively messy, such as a tree area, or the overall grayscale is relatively consistent, such as a roof area, the possibility that the first maximum value and the second maximum value in the grayscale level histogram corresponding to the local neighborhood of the pixel to be selected are significantly different is smaller, and the background information is relatively flat or messy as a whole. Therefore, the quality of the corresponding fitting straight line in the local neighborhood of the pixel to be selected is poor, thereby making the grayscale deviation and smoothing weight of the pixel to be selected smaller, and the structural prominence value of the pixel to be selected smaller.

[0064] According to the grayscale difference in the local neighborhood of each candidate pixel in the local window of each feature point, the structural prominence of each candidate pixel is obtained, reflecting the structural attribute of the candidate pixel. However, due to the high sampling position of remote sensing images, some pixels with pseudo structural attributes will appear in them. For example, a vehicle driving on the side of the street may identify a pixel in the boundary of the vehicle as a structural attribute pixel. However, when shooting optical remote sensing images and SAR remote sensing images, the position of the vehicle changes. At this time, the pseudo structural attribute pixels corresponding to the vehicle will interfere with the matching of feature points.

[0065] Based on the above analysis, for all the candidate pixels in the local window of each feature point, the structural prominence of each candidate pixel is obtained, and a screening threshold is set. In this embodiment, the screening threshold is ,in, The maximum value of the structural prominence of all candidate pixels in the local window representing the feature point, that is, 50% of the maximum structural prominence in the local window of the feature point is used as the screening threshold, and the pixels in the local window of the feature point whose structural prominence is greater than or equal to the screening threshold are marked as potential structural pixels, and the remaining pixels are marked as background pixels.

[0066] It should be noted that the implementer can set the percentage in the calculation screening threshold according to the actual situation, that is, the value in this embodiment is 50%, and there is no limitation for comparison in this embodiment.

[0067] All potential structural pixels within the local window of any feature point are combined into a potential structural pixel set, which includes real structural attribute pixels, such as streets and houses, and pseudo structural attribute pixels, such as vehicle edges. There are certain differences between the two, which are manifested as follows: since the overall continuity of pixels with structural attributes such as streets and houses is strong, and the whole presents the same grayscale information, the distribution of pseudo structural attribute pixels of vehicles is isolated and discontinuous, and due to the difference in the color of the vehicle itself, the grayscale of the pseudo structural attribute pixels is quite different from the overall.

[0068] The spatial coordinates of all pixels in the potential structural pixel set based on any feature point are used as the input of the DBSCAN algorithm, and the output is each cluster. The cluster containing the most potential structural pixels is marked as the real structural attribute cluster, and the remaining clusters are marked as clusters to be screened. In this embodiment, the hyperparameters of the DBSCAN algorithm are eps=3, min_samples=5, which can be set by the implementer according to the actual situation. The DBSCAN algorithm is an existing well-known technology. The implementer can choose other existing feasible clustering algorithms according to the actual situation, and this embodiment does not limit it here.

[0069] For each cluster to be screened, calculate the spatial grayscale deviation of each pixel in it. The specific calculation method is:

[0070] ; E represents the spatial grayscale deviation of any pixel in the cluster to be screened, I represents the grayscale value of any pixel, It represents the grayscale mean of all pixels in the real structural attribute cluster corresponding to the same feature point of the cluster to be screened, recorded as the second mean. represents the Euclidean distance between any pixel point and the cluster center of the real structural attribute cluster of the same feature point. Denoted as the candidate difference, Recorded as the distance to be selected.

[0071] By clustering all the pixels in the potential structural pixel set, the cluster containing the largest amount of data is taken as the real structural attribute cluster. For a single pixel in the cluster to be screened, if the grayscale value of the pixel deviates greatly from the grayscale value of the real structural attribute cluster and is far away from the cluster center of the real structural attribute cluster, the pixel is more likely to be a pseudo-structural attribute pixel, and the larger the value of the spatial grayscale deviation of the pixel is.

[0072] For each feature point, the spatial grayscale deviation of all pixels in the cluster to be screened is obtained, and the segmentation threshold of all the spatial grayscale deviations corresponding to each feature point is obtained using the Otsu threshold method. Pixels less than the segmentation threshold are marked as real structural attribute pixels, recorded as structural attribute pixels, and pixels greater than or equal to the segmentation threshold are marked as background pixels. It should be noted that all pixels in the real structural attribute cluster are also marked as real structural attribute pixels, recorded as structural attribute pixels. The flowchart for obtaining structural attribute pixels is shown in the figure. Figure 2 shown.

[0073] At this point, all the real structural attribute pixels and background pixels of each feature point are obtained, and all the real structural attribute pixels of each feature point are grouped into a structural set, and all the background pixels of each feature point are grouped into a background set. In this embodiment, the length of the feature descriptor of each feature point is set to Q=256, and the point pair is selected as follows: randomly select one of the pixels from the structural set as the first endpoint, and randomly select a pixel from the background set as the second endpoint, and the first endpoint and the second endpoint form a point pair. If the gray value of the first endpoint is greater than the second endpoint, it is marked as 1, otherwise, it is marked as 0. Randomly select Q groups of point pairs, and obtain the mark values ​​of the Q groups of point pairs in turn, thereby forming the feature descriptor V of each feature point as a binary string of length Q, and the descriptor V represents the structural information of the feature point. Among them, the construction of the feature descriptor of the feature point is an existing well-known technology, and this embodiment will not be described in detail here.

[0074] At this point, the feature descriptor of each feature point in the optical remote sensing image can be obtained. For the SAR remote sensing image, the same acquisition method as that of the optical remote sensing image is adopted to obtain the feature descriptor of each feature point in the SAR remote sensing image.

[0075] In this embodiment, a brute force matching algorithm based on cross-checking is adopted to match the optical remote sensing image and the SAR remote sensing image based on the descriptors of the feature points in the optical remote sensing image and the SAR remote sensing image, thereby obtaining a set of matching point pairs in the optical remote sensing image and the SAR remote sensing image, and completing the matching of the optical remote sensing image and the SAR remote sensing image. The brute force matching algorithm based on cross-checking is an existing well-known technology, and the implementer may adopt other existing feasible remote sensing image matching algorithms, which are not limited in this embodiment.

[0076] The matching results of the optical remote sensing image and the SAR remote sensing image in this embodiment are as follows: Figure 3 As shown in Figure 2, the matching results of the optical remote sensing image and the SAR remote sensing image obtained using the traditional ORB algorithm are as follows: Figure 4 As shown, in Figure 3 and Figure 4 In the figure, the left side is optical remote sensing images, and the right side is SAR remote sensing images; Figure 3 and Figure 4 It can be seen from the comparison that this embodiment has a more accurate remote sensing image matching result.

[0077] It should be noted that the above sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0078] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0079] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Modifications to the technical solutions recorded in the aforementioned embodiments, or equivalent replacement of some of the technical features therein, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A remote sensing image matching method for complex environments, characterized in that: The method comprises the following steps: Acquire multiple remote sensing images of the same target area; select multiple feature points in each remote sensing image; Each pixel point in the local window of each feature point is recorded as a candidate pixel point; the grayscale difference of the pixel points in the local neighborhood of each candidate pixel point is analyzed to determine the grayscale deviation of each candidate pixel point; based on the magnitude relationship between the grayscale difference of each pixel point in the local neighborhood of each candidate pixel point and the adjacent pixel points, each significant pixel point in the local neighborhood of each candidate pixel point is obtained; Based on the position coordinates of all significant pixels on the remote sensing image, a fitting straight line is obtained by fitting; the distance relationship between all the significant pixels of each candidate pixel and the fitting straight line is analyzed to determine the smoothing weight of each candidate pixel; Combining the grayscale deviation with the smoothing weight, the structural prominence of each candidate pixel is obtained; for each feature point, the candidate pixel whose structural prominence in the local window is greater than or equal to a preset percentage of the maximum value of all the structural prominences in the local window is recorded as a potential structural pixel, and the potential structural pixel points in the local window are clustered to obtain the real structural attribute cluster and each cluster to be screened; Analyze the difference between the grayscale values ​​of the pixels in each cluster to be screened and the real structural attribute cluster, as well as the distance relationship between the pixels in each cluster to be screened and the real structural attribute cluster, determine the spatial grayscale deviation of each pixel in each cluster to be screened, and obtain each structural attribute pixel of the local window of each feature point in combination with the spatial grayscale deviation and the real structural attribute cluster; determine each point pair of each feature point based on the structural attribute pixel point, mark each point pair to obtain the feature descriptor of each feature point, and use the image matching algorithm to obtain the matching results of the multiple remote sensing images; The structural prominence is the product of the grayscale deviation and the smoothing weight; The process of determining the real structural attribute cluster and each cluster to be screened is as follows: The real structure attribute cluster is the cluster containing the most potential structure pixels, and the clusters to be screened are the remaining clusters except the real structure attribute cluster; The process of determining the spatial grayscale deviation is as follows: For the local window of each feature point, the mean of the grayscale values ​​of all pixels in the real structural attribute cluster is calculated, which is recorded as the second mean. The difference between the grayscale value of each pixel in each cluster to be screened and the second mean is calculated, which is recorded as the selected difference. The metric distance between each pixel in each cluster to be screened and the cluster center of the real structural attribute cluster is calculated, which is recorded as the selected distance. The spatial grayscale deviation of each pixel in each cluster to be screened is the product of the selected difference and the selected distance. The process of determining each structural attribute pixel point is as follows: Threshold segmentation is performed on the spatial grayscale deviation of all pixels in the clusters to be screened, and the pixels in the clusters to be screened whose spatial grayscale deviation is less than the segmentation threshold and all the pixels in the real structural attribute cluster are determined as structural attribute pixels; The step of determining each point pair of each feature point based on the structural attribute pixel point comprises: The pixels other than the structural attribute pixels in the local window of each feature point are recorded as background pixels. Any structural attribute pixel and any background pixel in the local window of each feature point form a point pair.

2. A remote sensing image matching method for complex environments as claimed in claim 1, characterized in that: The grayscale histograms of all pixels in the local neighborhood of each candidate pixel are fitted to obtain a fitting curve, and the grayscale deviation is the difference between the grayscale level of the maximum maximum value and the grayscale level of the second maximum maximum value on the fitting curve.

3. The remote sensing image matching method for complex environments as claimed in claim 1, characterized in that: The process of determining each significant pixel point is as follows: The mean of the grayscale difference between each pixel and the adjacent pixel in the local neighborhood of each candidate pixel is calculated, the mean values ​​of all the pixels in the local neighborhood of each candidate pixel are arranged in descending order, and a preset number of pixels after the arrangement are used as the significant pixels.

4. The remote sensing image matching method for complex environments as claimed in claim 1, characterized in that: The determination process of the smoothing weight is: The mean of the metric distances between all the significant pixels of each candidate pixel and the fitting straight line is calculated and recorded as the first mean, and the smoothing weight is the reciprocal of the sum of the first mean and a preset value greater than 0.

5. The remote sensing image matching method for complex environments as claimed in claim 1, characterized in that: The process of determining the feature descriptor of each feature point is as follows: If the grayscale value of the structural attribute pixel in each point pair is greater than the grayscale value of the background pixel, the point pair is marked as 1, otherwise, it is marked as 0. The marking values ​​of all point pairs of each feature point form the feature descriptor of each feature point.

Images