Automatic registration method, device and system for remote sensing image data

Abstract

The invention relates to the technical field of image data processing, in particular to an automatic registration method, device and system for remote sensing image data, and the method comprises the steps: obtaining remote sensing images of a to-be-collected region at different moments; any two remote sensing images to be registered are analyzed, and key points in the two remote sensing images to be registered are obtained; obtaining reliable matching point pairs in the two to-be-registered remote sensing images according to the local network structure corresponding to each key point in the two to-be-registered remote sensing images and the brightness change of the key point corresponding to the local network structure; and matching all reliable matching point pairs in the two remote sensing images to be registered to obtain a registered target remote sensing image. The accuracy of remote sensing image registration can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and in particular to an automatic registration method, device and system for remote sensing image data. Background Art

[0002] Image registration is the process of matching and superimposing two or more images acquired at different times, using different sensors (imaging devices) or under different conditions (such as illumination, camera position and angle, etc.).

[0003] Remote sensing images are crucial for other environmental monitoring such as river leakage. By comparing remote sensing images acquired at different times, changes in river leakage can be identified, natural disasters can be monitored, etc.

[0004] However, when remote sensing images are automatically registered, due to the existence of self-similar structures in the remote sensing images, for example, there are large areas of farmland in the images, and a certain feature in the large area of ​​farmland appears multiple times in the remote sensing image in a similar pattern. When the matching algorithm selects features in the self-similar area, similar features in different locations may be mistakenly matched as the same features, resulting in deviations in the remote sensing image registration, which is not conducive to subsequent environmental monitoring based on the registered remote sensing images. Summary of the invention

[0005] In order to solve the technical problem of deviation in remote sensing image registration, the purpose of the present invention is to provide a method, device and system for automatic registration of remote sensing image data. The technical solution adopted is as follows: In a first aspect of the present disclosure, a method for automatic registration of remote sensing image data is provided, the method comprising: Obtain remote sensing images of the area to be collected at different times; Analyze any two remote sensing images to be registered to obtain key points in the two remote sensing images to be registered; Acquire reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure; All the reliable matching point pairs in the two remote sensing images to be registered are matched to obtain the registered target remote sensing image.

[0006] In one embodiment, obtaining reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure includes: Obtaining matching point pairs among key points of a source image and key points of a target image; the source image and the target image are the two remote sensing images to be registered; Obtaining a first stable matching point pair likelihood value for each pair of matching point pairs; Acquire a first target matching point pair whose likelihood value of the first stable matching point pair is greater than a first preset threshold; Acquire a first local network structure corresponding to each remaining key point of the source image; the remaining key points in the source image are the key points remaining after removing the key points corresponding to the first target matching point pair in the source image; Acquire a second local network structure corresponding to each remaining key point of the target image; the remaining key points in the target image are the key points remaining after removing the key points corresponding to the first target matching point pair in the target image; Obtaining reliable matching points of each remaining key point in the source image among the remaining key points in the target image according to the first local network structure, the second local network structure, the brightness of the first local network structure, and the brightness of the second local network structure; The reliable matching point pair is acquired according to the first target matching point pair and the second target matching point pair, wherein the second target matching point pair is composed of the remaining key points in the source image and the corresponding reliable matching points.

[0007] In one embodiment, obtaining matching point pairs between key points of the source image and key points of the target image includes: Obtaining a first descriptor of each key point of the source image; Obtaining a second descriptor of each key point in the target image; For each of the first descriptors of the key points of the source image, perform the following steps: Obtaining the distance between the current first descriptor and each of the second descriptors; Get the second descriptor of the target corresponding to the minimum distance; The key point of the source image corresponding to the current first descriptor and the key point in the target image corresponding to the target second descriptor form a pair of matching points.

[0008] In one embodiment, obtaining the first stable matching point pair likelihood value of each pair of matching point pairs includes: For each pair of matching points, perform the following steps: Obtain the key point distribution density within the first preset neighborhood corresponding to the current key point; Obtaining the key point distribution density in a second preset neighborhood corresponding to an initial nearest neighbor key point; the initial nearest neighbor key point and the current key point are a pair of matching point pairs; the current key point is a key point in the source image, and the initial nearest neighbor key point is a key point in the target image; According to the key point distribution density in the first preset neighborhood and the key point distribution density in the second preset neighborhood, obtaining the feature representation credibility of the current key point and the initial nearest neighbor key point; Obtain the nearest neighbor key point and the next nearest neighbor key point of the current key point in the target image; Respectively obtain the relative distances between the current key point and the initial nearest neighbor key point, the nearest neighbor key point, and the next nearest neighbor key point; Respectively obtain the angle values ​​corresponding to the initial nearest neighbor key point and the nearest neighbor key point in the target image; The first stable matching point pair possibility value between the current key point and the initial nearest neighbor key point is obtained according to the feature representation credibility, the relative distance and the angle value.

[0009] In one embodiment, The obtaining of a first local network structure corresponding to each remaining key point of the source image includes: Perform the following steps for each of the remaining key points of the source image: Acquire a preset number of first remaining key points that are closest to the current remaining key point in the source image from among all the remaining key points in the source image; Connecting the preset number of first remaining key points in sequence to obtain the first local network structure corresponding to the current remaining key points in the source image, wherein the current remaining key points in the source image are source key points of the first local network structure, and the preset number of first remaining key points are edge key points of the first local network structure; The obtaining of a second local network structure corresponding to each remaining key point of the target image includes: For each of the remaining key points of the target image, perform the following steps: Acquire, from all the remaining key points of the target image, the preset number of second remaining key points that are closest to the current remaining key point in the target image; Connect the preset number of second remaining key points in sequence to obtain the second local network structure corresponding to the current remaining key points in the target image, wherein the current remaining key points in the target image are the source key points of the second local network structure, and the preset number of second remaining key points are the edge key points of the second local network structure.

[0010] In one embodiment, obtaining reliable matching points of each remaining key point in the source image among the remaining key points in the target image according to the first local network structure, the second local network structure, the brightness of the first local network structure, and the brightness of the second local network structure includes: Perform the following steps for each of the first local network structures: Calculating an initial similarity between the current first local network structure and each of the second local network structures; Obtain the target second local network structure corresponding to the maximum initial similarity; Obtaining a final structural similarity between the current first partial network structure and the target second partial network structure; According to the brightness value of the current first local network structure and the brightness value of the target second local network structure, obtaining the ground vegetation color similarity between the current first local network structure and the target second local network structure; According to the final structural similarity and the ground vegetation color similarity, obtaining a second stable matching point pair possibility value between the source key point corresponding to the current first local network structure and the source key point in the target second local network structure; When the likelihood value of the second stable matching point pair is greater than a second preset threshold, the source key point in the target second local network structure is determined to be a reliable matching point of the source key point corresponding to the current first local network structure.

[0011] In one embodiment, obtaining the final structural similarity between the current first partial network structure and the target second partial network structure includes: Acquire a first distance between a source key point of the current first local network structure and each edge key point of the current first local network structure; Acquire a second distance between a source key point of the target second partial network structure and each of the edge key points of the target second partial network structure; Acquire a first angle between a source key point of the current first partial network structure and a line connecting each of the edge key points of the current first partial network structure and a horizontal line; Acquire a second angle between a source key point of the target second partial network structure and a line connecting each of the edge key points of the target second partial network structure and a horizontal line; According to the first distance, the second distance, the first angle, and the second angle, a final structural similarity between the current first local network structure and the target second local network structure is obtained.

[0012] In one embodiment, obtaining the ground vegetation color similarity between the current first local network structure and the target second local network structure according to the brightness value of the current first local network structure and the brightness value of the target second local network structure includes: Obtaining a first grayscale value of each edge key point of the current first local network structure; Acquire a second grayscale value of each edge key point of the target second local network structure; The ground vegetation color similarity is acquired according to all the first grayscale values ​​and all the second grayscale values.

[0013] In a second aspect of the present disclosure, there is provided an automatic registration device for remote sensing image data, the device comprising: The first acquisition module is used to acquire remote sensing images of the area to be collected at different times; A second acquisition module is used to analyze any two remote sensing images to be registered, and obtain key points in the two remote sensing images to be registered; A third acquisition module is used to acquire reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure; The matching module is used to match all the reliable matching point pairs in the two remote sensing images to be registered to obtain the target remote sensing image after registration.

[0014] In a third aspect of the present disclosure, there is provided an automatic registration system for remote sensing image data, the system comprising a drone and the automatic registration device for remote sensing image data as described in the second aspect; The drone is used to capture remote sensing images of the area to be collected at different times, and send the remote sensing images of the collection area at different times to the first acquisition module; The first acquisition module is used to acquire remote sensing images of the area to be collected at different times; The second acquisition module is used to analyze any two remote sensing images to be registered, and obtain key points in the two remote sensing images to be registered; The third acquisition module is used to acquire reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure; The matching module is used to match all the reliable matching point pairs in the two remote sensing images to be registered to obtain the registered target remote sensing image.

[0015] The present invention has the following beneficial effects: In the present invention, remote sensing images of a region to be collected at different times are obtained; any two remote sensing images to be registered are analyzed to obtain key points in the two remote sensing images to be registered; reliable matching point pairs in the two remote sensing images to be registered are obtained according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure; all reliable matching point pairs in the two remote sensing images to be registered are matched to obtain the target remote sensing image after registration. By obtaining reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure, the error matching problem caused by the self-similar structure can be effectively alleviated, and the accuracy of remote sensing image registration can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention 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 invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A process for an automatic registration method for remote sensing image data provided by an embodiment of the present invention Figure 1 ; Figure 2 An embodiment of the present invention provides a method for obtaining a remote sensing image of a region to be collected at a certain time by aerial photography; Figure 3 A schematic diagram of key points in a source image provided by an embodiment of the present invention; Figure 4 A schematic diagram of key points in a target image provided by an embodiment of the present invention; Figure 5 A schematic diagram of a target remote sensing image after registration provided by an embodiment of the present invention; Figure 6A process for an automatic registration method for remote sensing image data provided by an embodiment of the present invention Figure 2 ; Figure 7 A structural block diagram of an automatic registration device for remote sensing image data provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the automatic registration method, device and system for remote sensing image data proposed by the present invention, 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.

[0019] 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 invention belongs.

[0020] The following is a detailed description of a method, device and system for automatic registration of remote sensing image data provided by the present invention in conjunction with the accompanying drawings.

[0021] See also Figure 1 , which shows a flow chart of an automatic registration method for remote sensing image data provided by an embodiment of the present invention, such as Figure 1 As shown, the method includes the following steps S101-S104: S101, acquiring remote sensing images of the area to be collected at different times.

[0022] Multiple remote sensing images of the area to be collected can be obtained through aerial photography.

[0023] Determine the area and scope of data acquisition, use drones equipped with optical cameras or multispectral sensors to collect data in the area to be collected, and then obtain remote sensing images of the area to be collected at a certain time through digital scanning. By scanning the area to be collected multiple times at different times, multiple remote sensing images of the area to be collected can be obtained.

[0024] Among them, multiple remote sensing images were taken from approximately the same viewpoint during the flight of the drone, and when shooting, try to choose a time with good weather conditions.

[0025] If the area and range of data acquisition are large, a fixed-altitude cruise flight method can be used to collect data multiple times to obtain all remote sensing images of the area to be collected. Figure 2An embodiment of the present invention provides a method for obtaining a remote sensing image of a region to be collected at a certain time by aerial photography, such as Figure 2 As shown, the remote sensing image of the area to be collected mainly includes a large area of ​​farmland and a river.

[0026] S102: Analyze any two remote sensing images to be registered, and obtain key points in the two remote sensing images to be registered.

[0027] Perform key point detection on remote sensing images to obtain all possible key points in remote sensing images.

[0028] The key point detection method in the SIFT algorithm can be used to detect the key points of each remote sensing image to obtain all possible key points in the remote sensing image.

[0029] S103, obtaining reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness changes of the key points corresponding to the local network structure.

[0030] In this step, the two remote sensing images to be registered are analyzed to determine all reliable matching point pairs with high reliability in the two remote sensing images to be registered.

[0031] In the following embodiments, in order to clearly illustrate the solutions in the present application, the source image and the target image are used as two remote sensing images to be registered for illustration, that is, in the following embodiments, all reliable matching point pairs with high reliability in the source image and the target image are obtained.

[0032] In one embodiment, Figure 6 As shown, step S103 includes the following sub-steps S1031-S1037: S1031, obtaining matching point pairs between key points of the source image and key points of the target image.

[0033] Remote sensing image registration is the process of matching and superimposing two or more images to align the pixel precision between the images. In the image registration process, image matching is the most core step. Therefore, the main purpose of this step is to obtain as many possible key points as possible and all possible matching point pairs in the two remote sensing images to be registered, so as to subsequently screen stable and reliable matching point pairs in the two remote sensing images to be registered, and then complete the automatic registration of the two remote sensing images to be registered.

[0034] By performing key point detection on the two remote sensing images to be registered, all possible key points in the two remote sensing images to be registered can be obtained, and then by matching the descriptors of the key points in any two remote sensing images to be registered, all possible matching point pairs in the source image and the target image can be obtained. Specifically, step S1031 includes the following sub-steps S10311-S10315: S10311. Obtain the first descriptor of the key point of each source image.

[0035] The first descriptor of the key point in each source image can be obtained through the SIFT algorithm, wherein the first descriptor represents the descriptor in the source image.

[0036] S10312. Obtain the second descriptor of each key point in the target image.

[0037] The second descriptor of the key point in the target image can be obtained through the SIFT algorithm, wherein the second descriptor represents the descriptor in the target image.

[0038] The following steps S10312-S10314 are performed for the first descriptor of each key point of the source image: S10313. Obtain the distance between the current first descriptor and each second descriptor.

[0039] The distance may be the Euclidean distance.

[0040] S10314. Obtain the second descriptor of the target corresponding to the minimum distance.

[0041] S10315. The key point of the source image corresponding to the current first descriptor and the key point in the target image corresponding to the target second descriptor are a pair of matching points.

[0042] The two key points with the closest distance between the descriptors in the source image and the target image are regarded as a possible pair of matching points in the source image and the target image.

[0043] Figure 3 The diagram in the figure shows the key points in the source image. Figure 4 The figure below is a schematic diagram of the key points in the target image. The key points detected are mainly the edge points and corner points of the farmland in the remote sensing image, as well as the bright spots in the dark area or dark spots in the bright area of ​​the river. However, the detected key points are not all stable and reliable key points. This step is mainly to obtain as many possible key points and possible matching point pairs in the remote sensing image as possible. The subsequent steps will further screen and determine stable and reliable matching point pairs.

[0044] At this point, through key point detection processing, all possible key points in the source image and the target image, as well as all possible matching point pairs in the source image and the target image are obtained.

[0045] S1032: Obtain a first stable matching point pair likelihood value for each pair of matching points.

[0046] The main purpose of this step is to preliminarily determine the partial matching point pairs with obvious features, stable and reliable in the source image and the target image, that is, to determine the stable and reliable key points. The stable and reliable key points in the source image and the target image have obvious feature differences in local areas. The self-similar structure is manifested as similar patterns appearing in multiple local areas in the source image and the target image. For example, there are large areas of farmland in the source image and the target image, and a feature in the large area of ​​farmland has similar patterns appearing multiple times in the source image and the target image. The self-similar structure is often distributed in areas with complex key points. If the distance difference between the nearest neighbor key point and the second nearest neighbor point corresponding to a key point in the source image in the target image is larger, it means that this key point in the source image and the nearest key point in the target image are likely to be a correct match, with obvious feature differences, and are least affected by the self-similar structure. Both are stable key points; if the distance difference is smaller, it means that the distance between this key point in the source image and the two key points in the target image is close, and the matching point may be mismatched.

[0047] In the following embodiments, the initial nearest neighbor key point and the current key point are a pair of matching point pairs, wherein the current key point is a key point in the source image, and the initial nearest neighbor key point is a key point in the target image. In one embodiment, step S1032 includes performing the following steps S10321-S10327 for each pair of matching point pairs: S10321. Obtain the key point distribution density within a first preset neighborhood corresponding to the current key point.

[0048] For the current key point in the source image, in the source image, a square range of 11*11 with the current key point as the center is used as the first preset neighborhood range of the current key point, and the key point distribution density within the first preset neighborhood range is obtained. In the present disclosure, the number of key points within the first preset neighborhood range can be used to characterize the key point distribution density within the first preset neighborhood range.

[0049] S10322. Obtain the key point distribution density in the second preset neighborhood corresponding to the initial nearest neighbor key point.

[0050] The first preset neighborhood and the second preset neighborhood have the same corresponding range.

[0051] Another key point of the matching point pair corresponding to the current key point is used as the initial nearest neighbor key point of the current key point. For the initial nearest neighbor key point in the target image, in the target image, a square range of 11*11 with the initial nearest neighbor key point as the center is used as the second preset neighborhood range of the initial nearest neighbor key point, and the key point distribution density within the second preset neighborhood range is obtained. In the present disclosure, the number of key points within the second preset neighborhood range can be used to characterize the key point distribution density within the second preset neighborhood range.

[0052] S10323. Obtain the feature representation credibility of the current key point and the initial nearest neighbor key point according to the key point distribution density in the first preset neighborhood and the key point distribution density in the second preset neighborhood.

[0053] The feature performance credibility of the current key point and the initial nearest neighbor key point can be obtained by the following formula: ; In the formula, The source image The key point distribution density of a key point in its first preset neighborhood; The source image The key point distribution density of the initial nearest neighbor key point corresponding to the key point in the target image within its second preset neighborhood range; It is an exponential function with a natural constant as the base, used to realize an inverse proportional relationship; The source image The reliability of the feature performance of the key point and the corresponding initial nearest neighbor key point in the target image. The key points represent the current key points of the source image in the above embodiment.

[0054] In the above formula, the first The smaller the key point distribution density is within the first preset neighborhood, the smaller the key point distribution density is. The key point distribution density of the initial nearest neighbor key point corresponding to the key point in the target image within its second preset neighborhood is the same as that of the first nearest neighbor key point in the source image. The closer the key point distribution density is within the first preset neighborhood, the more The better the sparse feature performance of the key points and the corresponding initial nearest neighbor key points in the target image, the less affected by the self-similar structure, and the higher the credibility of the feature performance.

[0055] After preliminary analysis of the above steps, the reliability of the feature performance of the current key point in the source image and the corresponding initial nearest neighbor key point in the target image is obtained. Considering that the remote sensing images are taken from approximately the same viewpoint during the flight of the drone, there will be a certain offset between the images. Therefore, it is necessary to combine the changes in the nearest neighbor distance and the next nearest neighbor distance to obtain stable and reliable matching point pairs.

[0056] S10324. Obtain the nearest neighbor key point and the next nearest neighbor key point of the current key point in the target image.

[0057] In this step, since the current key point in the source image and the initial nearest neighbor key point in the target image are a pair of matching points, the nearest neighbor key point and the next nearest neighbor key point corresponding to the initial nearest neighbor key point can be obtained in the target image. The nearest neighbor key point and the next nearest neighbor key point corresponding to the initial nearest neighbor key point obtained here represent the nearest neighbor key point and the next nearest neighbor key point of the current key point in the target image in the above embodiment.

[0058] S10325. Obtain the relative distances between the current key point and the initial nearest neighbor key point, the nearest neighbor key point, and the next nearest neighbor key point respectively.

[0059] S10326. Obtain the angle values ​​corresponding to the initial nearest neighbor key point and the nearest neighbor key point in the target image respectively.

[0060] S10327. Obtain the probability value of the first stable matching point pair between the current key point and the initial nearest neighbor key point according to the feature representation credibility, relative distance and angle value.

[0061] The probability value of the first stable matching point pair is obtained by the following formula: ; In the formula, The source image The credibility of the feature representation of the key points and the corresponding initial nearest neighbor key points in the target image; The source image The relative distance between the key point and the corresponding initial nearest neighbor key point in the target image. The relative distance is specifically The Euclidean distance between the coordinates of a key point and the coordinates of the corresponding initial nearest neighbor key point; The source image The relative distance between a key point and the corresponding nearest neighbor key point in the target image; is the relative distance parameter, is an angle parameter, You can take 0.6. It can be set to 0.4. The smaller the relative distance, the more it indicates a correct match. Therefore, the relative distance parameter can be set to 0.6. There will be a certain offset between the images. The smaller the offset, the better. However, the correct match is mainly determined by the relative distance, which is less important. Therefore, the angle parameter can be set to 0.4. is a hyperparameter to prevent the denominator from being 0. It can be 0.1; The source image The angle value corresponding to the initial nearest neighbor key point in the target image is the angle value between the line segment formed by the initial nearest neighbor key point and the lower left corner of the target image and the horizontal line. The source image The angle value of the nearest neighbor key point in the target image corresponding to the key point in the target image; the angle value of the nearest neighbor key point in the target image is the angle between the line segment formed by the nearest neighbor key point and the lower left corner of the target image and the horizontal line; The source image The relative distance between a key point and the corresponding next nearest key point in the target image; The source image The probability value of the first stable matching point pair between the key point and the corresponding initial nearest neighbor key point in the target image can be used to determine the first stable matching point pair in the source image. Whether the key point and the corresponding initial nearest neighbor key point in the target image are a stable and reliable matching point pair.

[0062] When the feature performance credibility is greater, it is least affected by the self-similar structure. The smaller the relative distance difference between the key point and the corresponding initial nearest neighbor key point and the nearest neighbor key point in the target image, the smaller the relative distance difference between the first key point and the target image. The smaller the angle difference between the first key point and the corresponding initial nearest neighbor key point and the nearest neighbor key point in the target image, the smaller the angle difference between the first key point and the corresponding initial nearest neighbor key point in the target image. The more likely the initial nearest neighbor key point in the target image corresponds to the correct key point, the more likely it is that the corresponding key point is correct. The relative distance between the key point and the corresponding initial nearest neighbor key point in the target image, the The greater the relative distance difference between the first key point and the corresponding second nearest key point in the target image, the greater the difference in the relative distance between the first key point and the corresponding second nearest key point in the target image. The more likely a key point is to be a correct match with the initial nearest neighbor key point in the target image, the more obvious the feature difference is, and the less affected by the self-similar structure, and both are stable key points.

[0063] S1033: Obtain a first target matching point pair whose probability value of a first stable matching point pair is greater than a first preset threshold.

[0064] The first preset threshold can be set to 0.8, and the matching point pairs whose likelihood values ​​of the first stable matching point pairs obtained in the above step are greater than the set first preset threshold are regarded as stable and reliable matching point pairs, otherwise they are regarded as unstable and reliable matching point pairs, wherein the stable and reliable matching point pairs represent the first target matching points.

[0065] In one implementable method, before comparison, the probability value of the first stable matching point pair obtained in the above step can also be linearly normalized to obtain the normalized probability, and the matching point pairs whose normalized probability is greater than the set first preset threshold are regarded as stable and reliable matching point pairs, otherwise they are regarded as unstable and unreliable matching point pairs.

[0066] At this point, by analyzing the changes in the nearest neighbor distance and the next nearest neighbor distance, some matching point pairs with higher reliability in the source image and the target image are preliminarily obtained.

[0067] Next, the remaining key points are analyzed to obtain the local network structure of the polygon formed by each remaining key point in the source image and the target image and the surrounding similar remaining key points. The similarity of the local network structure chain codes of the polygons formed in the source image and the target image, as well as the brightness changes and density distribution changes of the remaining key points in the local network structure of the polygon are analyzed to determine more reliable matching point pairs in the source image and the target image.

[0068] For automatic registration of remote sensing images, it is hoped to determine as many, relatively accurate and evenly distributed matching point pairs as possible. The above analysis preliminarily obtains a small number of stable and reliable matching point pairs in the source image and the target image. The self-similar structures in the source image and the target image often exist in areas with complex local key points. In order to achieve better automatic registration of the source image and the target image, it is also necessary to verify the remaining possible stable key points to expand more stable matching point pairs.

[0069] The self-similar structures in the source and target images often have similar patterns in multiple local areas, which makes it difficult to determine the correct correspondence when selecting matching point pairs, resulting in incorrect matching. Therefore, if the matching point pairs are determined only by analyzing the performance of the local areas where the remaining key points are located, a large deviation will occur.

[0070] The stable remaining key points often form an approximately fixed local network structure. Therefore, the similarity of the local network structure can be analyzed to reduce the influence of the self-similar structure in the image. At the same time, if the color of the remaining key point area corresponding to the local network structure on the remote sensing image is similar, it means that the corresponding actual ground performance is similar within a certain neighborhood range, which can also reduce the influence of the self-similar structure. By analyzing the density distribution, if the distribution is relatively uniform, it means that the two remaining key points are a pair of correct matching points, thereby reducing the interference of self-similar matching points in the image, and expanding more stable matching point pairs to achieve automatic registration.

[0071] The key points in the unstable and reliable matching point pairs obtained above are used as the remaining key points in the source image and the target image. That is, the remaining key points in the source image are the key points remaining after removing the key points corresponding to the first target matching point pair in the source image, and the remaining key points in the target image are the key points remaining after removing the key points corresponding to the first target matching point pair in the target image. These remaining key points are further analyzed below.

[0072] S1034. Obtain a first local network structure corresponding to each remaining key point of the source image.

[0073] In one embodiment, step S1034 includes performing the following steps S10341-S10342 on each remaining key point in the source image: S10341. From all remaining key points of the source image, obtain a preset number of first remaining key points that are closest to the current remaining key point in the source image.

[0074] S10342. Connect a preset number of first remaining key points in sequence to obtain a first local network structure corresponding to the current remaining key points of the source image, wherein the current remaining key points of the source image are source key points of the first local network structure, and the preset number of first remaining key points are edge key points of the first local network structure.

[0075] For example, the preset number is 6. For any remaining key point in any source image, the 6 (other values ​​can also be set) first remaining key points closest to the remaining key point are obtained, and the 6 first remaining key points closest to the remaining key point are connected in sequence to obtain the first local network structure of the remaining key point.

[0076] S1035, obtaining a second local network structure corresponding to each remaining key point in the target image; In one embodiment, step S1035 includes performing the following steps S10351-S10352 on each remaining key point in the target image: S10351. From all remaining key points in the target image, obtain a preset number of second remaining key points that are closest to the current remaining key points of the target image.

[0077] S10352. Connect a preset number of second remaining key points in sequence to obtain a second local network structure corresponding to the current remaining key points of the target image, wherein the current remaining key points of the target image are source key points of the second local network structure, and the preset number of second remaining key points are edge key points of the second local network structure.

[0078] For example, the preset number is 6. For any remaining key point in any target image, the 6 (other values ​​can also be set) second remaining key points closest to the remaining key point are obtained, and the 6 second remaining key points closest to the remaining key point are connected in sequence to obtain the second local network structure of the remaining key point.

[0079] S1036. Obtain reliable matching points of each remaining key point in the source image among the remaining key points in the target image according to the first local network structure, the second local network structure, the brightness of the first local network structure, and the brightness of the second local network structure.

[0080] In one embodiment, step S1036 includes executing the following steps S10361-S10366 for each first local network structure: S10361. Calculate the initial similarity between the current first local network structure and each second local network structure.

[0081] The chain code of the current first local network structure is obtained, wherein the starting point of the chain code of the current first local network structure is obtained by circulating counterclockwise from an edge key point closest to a source key point of the current first local network structure.

[0082] The chain code of the second local network structure is obtained, wherein the starting point of the chain code of the second local network structure is obtained by starting from an edge key point closest to a source key point of the second local network structure and circulating counterclockwise.

[0083] The initial similarity between the first local network structure and the second local network structure is obtained by the following formula: ; In the formula, is the number of remaining key points in the target image; The source image The chain code sequence vector of the first local network structure of the remaining key points, the specific method of obtaining the chain code sequence vector is: The chain codes of the first local network structure of the remaining key points are arranged from the starting point to obtain the corresponding chain code sequence, and the chain code sequence is used as a row vector, that is, a chain code sequence vector; The target image The chain code sequence vector of the second local network structure of the remaining key points; is the cosine function, Represents the cosine similarity of two chain code sequence vectors; is the sigmoid function, used for normalization; The source image The first local network structure of the remaining key points is The initial similarity of the second local network structure of the remaining key points.

[0084] Among them, the source image The first local network structure of the remaining key points represents the current first local network structure; The remaining key points represent the source key points of the current first local network structure; the The remaining key points represent the corresponding source key points of the second local network structure.

[0085] S10362. Obtain the target second local network structure corresponding to the maximum initial similarity.

[0086] For each first local network structure, a target second local network structure corresponding to the maximum initial similarity is found, wherein the first local network structure and the target second local network structure corresponding to the corresponding maximum initial similarity are preliminarily similar.

[0087] S10363. Obtain the final structural similarity between the current first partial network structure and the target second partial network structure.

[0088] If the chain code sequence vector of the first local network structure of the remaining key points in the source image is very similar to the chain code sequence vector of the second local network structure of a remaining key point in the target image, it means that the source key points corresponding to the two local network structures are more likely to be a pair of correct matches, and the initial similarity of the two local network structures is better.

[0089] Considering that there may be multiple local network structures that are similar, which may lead to errors in the similarity judgment of local network structures, the distribution relationship of key points on the local network structure is analyzed here, and combined with the initial similarity, the final structural similarity of the local network structure of key points in the source image and the target image is determined.

[0090] In one embodiment, step S10363 includes the following sub-steps S103631-S103635: S103631. Obtain a first distance between a source key point of the current first local network structure and each edge key point of the current first local network structure.

[0091] S103632. Obtain a second distance between a source key point of the target second partial network structure and each edge key point of the target second partial network structure.

[0092] S103633. Obtain a first angle between a source key point of the current first local network structure and a line connecting each edge key point of the current first local network structure and a horizontal line.

[0093] S103634. Obtain a second angle between the source key point of the target second partial network structure and the connecting line between each edge key point of the target second partial network structure and the horizontal line.

[0094] S103635. Obtain a final structural similarity between the current first local network structure and the target second local network structure according to the first distance, the second distance, the first angle, and the second angle.

[0095] The final structural similarity between the current first local network structure and the target second local network structure can be obtained by the following formula: ; In the formula, is the number of edge key points on any local network structure, which can be the number of edge key points on the current first local network structure, or the number of edge key points on the target second local network structure. The source image The first local network structure of the remaining key points The remaining key points are The first distance of the remaining key points; is the first The first local network structure of the remaining key points has the highest initial similarity to the target second local network structure. The second distance of the remaining key points from the source key points of the target second local network structure; The source image The first local network structure of the remaining key points The remaining key points and The first angle between the line connecting the remaining key points and the horizontal line; is the first The first local network structure of the remaining key points has the highest initial similarity to the target second local network structure. The second angle value between the line connecting the remaining key points and the source key points of the second local network structure of the target and the horizontal line; The source image The first local network structure of the remaining key points is The initial similarity of the second local network structure of the remaining key points is assumed here. The second local network structure of the remaining key points is The first local network structure of the remaining key points is initially similar to that of the first A second local network structure with a similar initial first local network structure for the remaining key points; It is an exponential function with a natural constant as the base, used to realize inverse proportional relationship and normalization; The source image The first local network structure of the remaining key points is The final structural similarity of the second local network structure of the remaining key points.

[0096] Among them, the source image The first local network structure of the remaining key points represents the current first local network structure. The remaining key points represent the source key points in the current first local network structure. The first local network structure of the remaining key points The remaining key points represent the edge key points in the current first local network structure; The second local network structure of the remaining key points represents the target second local network structure corresponding to the current first local network structure; The remaining key points represent the source key points in the second local network structure of the target; the The remaining key points represent the edge key points in the second local network structure of the target.

[0097] When the corresponding local network structures in the source image and the target image are initially similar, if the differences in the distance distribution and angle distribution of the local network structures are small, it means that the two local network structures are truly similar, and the two remaining key points corresponding to the local network structures are more likely to be a pair of stable and reliable matching points.

[0098] Next, we analyze the brightness changes of the remaining key points in the two initially similar local network structures.

[0099] S10364. Obtain the ground vegetation color similarity between the current first local network structure and the target second local network structure according to the brightness value of the current first local network structure and the brightness value of the target second local network structure.

[0100] In one embodiment, step S10364 includes the following sub-steps S103641-S103643: S103641. Obtain a first grayscale value of each edge key point of the current first local network structure; S103642. Obtain a second grayscale value of each edge key point of the target second local network structure; S103643. Obtain ground vegetation color similarity according to all first grayscale values ​​and all second grayscale values.

[0101] Specifically, the ground vegetation color similarity is obtained according to the following formula: ; In the formula, is the number of edge key points on any local network structure, which may be the number of edge key points on the current first local network structure, or the number of edge key points on the target second local network structure; The source image The first local network structure of the remaining key points Grayscale values ​​of the remaining key points; is the first The first local network structure of the remaining key points is similar to the target second local network structure. Grayscale values ​​of the remaining key points; The source image The first local network structure of the remaining key points is The ground vegetation color similarity of the target second local network structure of the remaining key points, where the ground vegetation color similarity represents the consistency of the ground vegetation color performance in the actual scene. Here, it is assumed that The target second local network structure of the remaining key points is The first local network structure of the remaining key points is initially similar.

[0102] If the corresponding remaining key points on the actual ground of two initially similar local network structures have the same color, that is, the grayscale values ​​of the corresponding remaining key points are similar, it further indicates that the two local network structures are truly similar, and the two remaining key points corresponding to the local network structures are more likely to be a pair of stable and reliable matching points.

[0103] S10365. According to the final structure similarity and the ground vegetation color similarity, obtain possible values ​​of second stable matching point pairs between the source key points corresponding to the current first local network structure and the source key points in the target second local network structure.

[0104] Combining the final structural similarity and ground vegetation color similarity, the possible value of the second stable matching point pair is obtained by the following formula: ; In the formula, The source image The first local network structure of the remaining key points is The final structural similarity of the target second local network structure of the remaining key points; The source image The first local network structure of the remaining key points is The ground vegetation color similarity of the second local network structure of the remaining key points; is a linear normalization function; The source image The remaining key points and the The possible values ​​of the second stable matching point pairs of the remaining key points.

[0105] S10366: When the possible value of the second stable matching point pair is greater than a second preset threshold, determine that the source key point in the target second partial network structure is a reliable matching point of the source key point corresponding to the current first partial network structure.

[0106] The ones that are greater than the second preset threshold are regarded as stable and reliable matching point pairs, otherwise they are regarded as unstable and reliable matching point pairs.

[0107] At this point, more stable and reliable matching point pairs are obtained by analyzing the similarity of the network structure chain codes of the remaining key points and the brightness changes of the remaining key points in the local network structure of the polygon.

[0108] S1037. Obtain a reliable matching point pair according to the first target matching point pair and the second target matching point pair, where the second target matching point pair is composed of the remaining key points in the source image and the corresponding reliable matching points.

[0109] S104: Match all reliable matching point pairs in the two remote sensing images to be registered to obtain a registered target remote sensing image.

[0110] Match all the reliable matching point pairs in the two remote sensing images to be registered to obtain the registered target remote sensing image. Figure 5 As shown, all stable and reliable matching point pairs in the source image and the target image are matched using the SIFT algorithm to obtain the registered target remote sensing image.

[0111] This application analyzes the changes in the nearest neighbor distance and the next nearest neighbor distance of the corresponding key points in the source image and the target image, and considers the distance between the corresponding two pairs of key points in the two images, and preliminarily obtains some matching point pairs with higher reliability in the remote sensing image, and analyzes the remaining key points to obtain the polygonal network structure composed of each key point in the remote sensing image and the surrounding similar key points, analyzes the similarity of the polygonal network structure chain code composed of the two images to be registered, and the brightness changes of the key points in the polygonal network structure, and expands more reliable matching point pairs, so that the remote sensing image can still achieve automatic registration well when there is a slight offset and self-similar structure image. Through this method, the error matching problem caused by the self-similar structure can be effectively reduced, and the accuracy of remote sensing image registration can be improved.

[0112] Figure 7 A structural block diagram of an automatic registration device for remote sensing image data provided by an embodiment of the present invention, such as Figure 7 As shown, the device includes: The first acquisition module 11 is used to acquire remote sensing images of the area to be collected at different times; A second acquisition module 12 is used to analyze any two remote sensing images to be registered, and obtain key points in the two remote sensing images to be registered; A third acquisition module 13 is used to acquire reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure; The matching module 14 is used to match all the reliable matching point pairs in the two remote sensing images to be registered to obtain a registered target remote sensing image.

[0113] In one embodiment, the third acquisition module is specifically used to: Obtaining matching point pairs among key points of a source image and key points of a target image; the source image and the target image are the two remote sensing images to be registered; Obtaining a first stable matching point pair likelihood value for each pair of matching point pairs; Acquire a first target matching point pair whose likelihood value of the first stable matching point pair is greater than a first preset threshold; Acquire a first local network structure corresponding to each remaining key point of the source image; the remaining key points in the source image are the key points remaining after removing the key points corresponding to the first target matching point pair in the source image; Acquire a second local network structure corresponding to each remaining key point of the target image; the remaining key points in the target image are the key points remaining after removing the key points corresponding to the first target matching point pair in the target image; Obtaining reliable matching points of each remaining key point in the source image among the remaining key points in the target image according to the first local network structure, the second local network structure, the brightness of the first local network structure, and the brightness of the second local network structure; The reliable matching point pair is acquired according to the first target matching point pair and the second target matching point pair, wherein the second target matching point pair is composed of the remaining key points in the source image and the corresponding reliable matching points.

[0114] In one embodiment, in the aspect of obtaining the matching point pairs between the key points of the source image and the key points of the target image, the third obtaining module is specifically used to: Obtaining a first descriptor of each key point of the source image; Obtaining a second descriptor of each key point in the target image; For each of the first descriptors of the key points of the source image, perform the following steps: Obtaining the distance between the current first descriptor and each of the second descriptors; Get the second descriptor of the target corresponding to the minimum distance; The key point of the source image corresponding to the current first descriptor and the key point in the target image corresponding to the target second descriptor form a pair of matching points.

[0115] In one embodiment, in the aspect of obtaining the first stable matching point pair likelihood value of each pair of matching point pairs, the third obtaining module is specifically configured to: For each pair of matching points, perform the following steps: Obtain the key point distribution density within the first preset neighborhood corresponding to the current key point; Obtaining the key point distribution density in a second preset neighborhood corresponding to an initial nearest neighbor key point; the initial nearest neighbor key point and the current key point are a pair of matching point pairs; the current key point is a key point in the source image, and the initial nearest neighbor key point is a key point in the target image; According to the key point distribution density in the first preset neighborhood and the key point distribution density in the second preset neighborhood, obtaining the feature representation credibility of the current key point and the initial nearest neighbor key point; Obtain the nearest neighbor key point and the next nearest neighbor key point of the current key point in the target image; Respectively obtain the relative distances between the current key point and the initial nearest neighbor key point, the nearest neighbor key point, and the next nearest neighbor key point; Respectively obtain the angle values ​​corresponding to the initial nearest neighbor key point and the nearest neighbor key point in the target image; The first stable matching point pair possibility value between the current key point and the initial nearest neighbor key point is obtained according to the feature representation credibility, the relative distance and the angle value.

[0116] In one embodiment, in the aspect of obtaining the first local network structure corresponding to each remaining key point of the source image, the third obtaining module is specifically used to: Perform the following steps for each of the remaining key points of the source image: Acquire a preset number of first remaining key points that are closest to the current remaining key point in the source image from among all the remaining key points in the source image; Connecting the preset number of first remaining key points in sequence to obtain the first local network structure corresponding to the current remaining key points in the source image, wherein the current remaining key points in the source image are source key points of the first local network structure, and the preset number of first remaining key points are edge key points of the first local network structure; In one embodiment, in the aspect of acquiring the second local network structure corresponding to each remaining key point of the target image, the third acquisition module is specifically used to: For each of the remaining key points of the target image, perform the following steps: Acquire, from all the remaining key points of the target image, the preset number of second remaining key points that are closest to the current remaining key point in the target image; Connect the preset number of second remaining key points in sequence to obtain the second local network structure corresponding to the current remaining key points in the target image, wherein the current remaining key points in the target image are the source key points of the second local network structure, and the preset number of second remaining key points are the edge key points of the second local network structure.

[0117] In one embodiment, in the aspect of acquiring reliable matching points of each remaining key point in the source image among the remaining key points in the target image according to the first local network structure, the second local network structure, the brightness of the first local network structure, and the brightness of the second local network structure, the third acquisition module is specifically used to: Perform the following steps for each of the first local network structures: Calculating an initial similarity between the current first local network structure and each of the second local network structures; Obtain the target second local network structure corresponding to the maximum initial similarity; Obtaining a final structural similarity between the current first partial network structure and the target second partial network structure; According to the brightness value of the current first local network structure and the brightness value of the target second local network structure, obtaining the ground vegetation color similarity between the current first local network structure and the target second local network structure; According to the final structural similarity and the ground vegetation color similarity, obtaining a second stable matching point pair possibility value between the source key point corresponding to the current first local network structure and the source key point in the target second local network structure; When the likelihood value of the second stable matching point pair is greater than a second preset threshold, the source key point in the target second local network structure is determined to be a reliable matching point of the source key point corresponding to the current first local network structure.

[0118] In one embodiment, in the aspect of obtaining the final structural similarity between the current first local network structure and the target second local network structure, the third obtaining module is specifically configured to: Acquire a first distance between a source key point of the current first local network structure and each edge key point of the current first local network structure; Acquire a second distance between a source key point of the target second partial network structure and each of the edge key points of the target second partial network structure; Acquire a first angle between a source key point of the current first partial network structure and a line connecting each of the edge key points of the current first partial network structure and a horizontal line; Acquire a second angle between a source key point of the target second partial network structure and a line connecting each of the edge key points of the target second partial network structure and a horizontal line; According to the first distance, the second distance, the first angle, and the second angle, a final structural similarity between the current first local network structure and the target second local network structure is obtained.

[0119] In one embodiment, in the aspect of acquiring the ground vegetation color similarity between the current first local network structure and the target second local network structure according to the brightness value of the current first local network structure and the brightness value of the target second local network structure, the third acquisition module is specifically used to: Obtaining a first grayscale value of each edge key point of the current first local network structure; Acquire a second grayscale value of each edge key point of the target second local network structure; The ground vegetation color similarity is acquired according to all the first grayscale values ​​and all the second grayscale values.

[0120] The present disclosure also provides an automatic registration system for remote sensing image data, the system comprising a drone and an automatic registration device for remote sensing image data as in the above-mentioned embodiment.

[0121] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. 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.

[0122] 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.

Claims

1. A method for automatic registration of remote sensing image data, characterized in that: The method comprises: Obtain remote sensing images of the area to be collected at different times; Analyze any two remote sensing images to be registered to obtain key points in the two remote sensing images to be registered; Acquire reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure; All the reliable matching point pairs in the two remote sensing images to be registered are matched to obtain the registered target remote sensing image.

2. The automatic registration method for remote sensing image data according to claim 1, characterized in that: The step of obtaining reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure comprises: Obtaining matching point pairs among key points of a source image and key points of a target image; the source image and the target image are the two remote sensing images to be registered; Obtaining a first stable matching point pair likelihood value for each pair of matching point pairs; Acquire a first target matching point pair whose likelihood value of the first stable matching point pair is greater than a first preset threshold; Acquire a first local network structure corresponding to each remaining key point of the source image; the remaining key points in the source image are the key points remaining after removing the key points corresponding to the first target matching point pair in the source image; Acquire a second local network structure corresponding to each remaining key point of the target image; the remaining key points in the target image are the key points remaining after removing the key points corresponding to the first target matching point pair in the target image; Obtaining reliable matching points of each remaining key point in the source image among the remaining key points in the target image according to the first local network structure, the second local network structure, the brightness of the first local network structure, and the brightness of the second local network structure; The reliable matching point pair is acquired according to the first target matching point pair and the second target matching point pair, wherein the second target matching point pair is composed of the remaining key points in the source image and the corresponding reliable matching points.

3. The automatic registration method for remote sensing image data according to claim 2, characterized in that: The step of obtaining matching point pairs between key points of the source image and key points of the target image includes: Obtaining a first descriptor of each key point of the source image; Obtaining a second descriptor of each key point in the target image; For each of the first descriptors of the key points of the source image, perform the following steps: Obtaining the distance between the current first descriptor and each of the second descriptors; Get the second descriptor of the target corresponding to the minimum distance; The key point of the source image corresponding to the current first descriptor and the key point in the target image corresponding to the target second descriptor form a pair of matching points.

4. The automatic registration method for remote sensing image data according to claim 2, characterized in that: The obtaining of the first stable matching point pair likelihood value of each pair of matching point pairs comprises: For each pair of matching points, perform the following steps: Obtain the key point distribution density within the first preset neighborhood corresponding to the current key point; Obtaining the key point distribution density in a second preset neighborhood corresponding to an initial nearest neighbor key point; the initial nearest neighbor key point and the current key point are a pair of matching point pairs; the current key point is a key point in the source image, and the initial nearest neighbor key point is a key point in the target image; According to the key point distribution density in the first preset neighborhood and the key point distribution density in the second preset neighborhood, obtaining the feature representation credibility of the current key point and the initial nearest neighbor key point; Obtain the nearest neighbor key point and the next nearest neighbor key point of the current key point in the target image; Respectively obtain the relative distances between the current key point and the initial nearest neighbor key point, the nearest neighbor key point, and the next nearest neighbor key point; Respectively obtain the initial nearest neighbor key point and the angle value corresponding to the nearest neighbor key point in the target image; The first stable matching point pair possibility value between the current key point and the initial nearest neighbor key point is obtained according to the feature representation credibility, the relative distance and the angle value.

5. The automatic registration method for remote sensing image data according to claim 2, characterized in that: The obtaining of a first local network structure corresponding to each remaining key point of the source image includes: Perform the following steps for each of the remaining key points of the source image: Acquire a preset number of first remaining key points that are closest to the current remaining key point in the source image from among all the remaining key points in the source image; Connecting the preset number of first remaining key points in sequence to obtain the first local network structure corresponding to the current remaining key points in the source image, wherein the current remaining key points in the source image are source key points of the first local network structure, and the preset number of first remaining key points are edge key points of the first local network structure; The obtaining of a second local network structure corresponding to each remaining key point of the target image includes: For each of the remaining key points of the target image, perform the following steps: Acquire, from all the remaining key points of the target image, the preset number of second remaining key points that are closest to the current remaining key point in the target image; Connect the preset number of second remaining key points in sequence to obtain the second local network structure corresponding to the current remaining key points in the target image, wherein the current remaining key points in the target image are the source key points of the second local network structure, and the preset number of second remaining key points are the edge key points of the second local network structure.

6. The automatic registration method for remote sensing image data according to claim 2, characterized in that: The step of obtaining reliable matching points of each remaining key point in the source image among the remaining key points in the target image according to the first local network structure, the second local network structure, the brightness of the first local network structure, and the brightness of the second local network structure includes: Perform the following steps for each of the first local network structures: Calculating an initial similarity between the current first local network structure and each of the second local network structures; Obtain the target second local network structure corresponding to the maximum initial similarity; Obtaining a final structural similarity between the current first partial network structure and the target second partial network structure; According to the brightness value of the current first local network structure and the brightness value of the target second local network structure, obtaining the ground vegetation color similarity between the current first local network structure and the target second local network structure; According to the final structural similarity and the ground vegetation color similarity, obtaining a second stable matching point pair possibility value between the source key point corresponding to the current first local network structure and the source key point in the target second local network structure; When the likelihood value of the second stable matching point pair is greater than a second preset threshold, the source key point in the target second local network structure is determined to be a reliable matching point of the source key point corresponding to the current first local network structure.

7. The automatic registration method for remote sensing image data according to claim 6, characterized in that: The obtaining of the final structural similarity between the current first partial network structure and the target second partial network structure includes: Acquire a first distance between a source key point of the current first partial network structure and each edge key point of the current first partial network structure; Acquire a second distance between a source key point of the target second partial network structure and each edge key point of the target second partial network structure; Acquire a first angle between a line connecting a source key point of the current first partial network structure and each edge key point of the current first partial network structure and a horizontal line; Acquire a second angle between a source key point of the target second partial network structure and a line connecting each edge key point of the target second partial network structure and a horizontal line; According to the first distance, the second distance, the first angle, and the second angle, a final structural similarity between the current first local network structure and the target second local network structure is obtained.

8. The automatic registration method for remote sensing image data according to claim 6, characterized in that: The obtaining, according to the brightness value of the current first local network structure and the brightness value of the target second local network structure, the ground vegetation color similarity between the current first local network structure and the target second local network structure includes: Obtaining a first grayscale value of each edge key point of the current first local network structure; Obtaining a second grayscale value of each edge key point of the target second local network structure; The ground vegetation color similarity is acquired according to all the first grayscale values ​​and all the second grayscale values.

9. An automatic registration device for remote sensing image data, characterized in that: The device comprises: The first acquisition module is used to acquire remote sensing images of the area to be collected at different times; A second acquisition module is used to analyze any two remote sensing images to be registered, and obtain key points in the two remote sensing images to be registered; A third acquisition module is used to acquire reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure; The matching module is used to match all the reliable matching point pairs in the two remote sensing images to be registered to obtain the target remote sensing image after registration.

10. An automatic registration system for remote sensing image data, characterized in that: The system comprises a drone and an automatic registration device for remote sensing image data as claimed in claim 9; The drone is used to capture remote sensing images of the area to be collected at different times, and send the remote sensing images of the collection area at different times to the first acquisition module; The first acquisition module is used to acquire remote sensing images of the area to be collected at different times; The second acquisition module is used to analyze any two remote sensing images to be registered, and obtain key points in the two remote sensing images to be registered; The third acquisition module is used to acquire reliable matching point pairs in the two remote sensing images to be registered according to the local network structure corresponding to each key point in the two remote sensing images to be registered and the brightness change of the key point corresponding to the local network structure; The matching module is used to match all the reliable matching point pairs in the two remote sensing images to be registered to obtain the registered target remote sensing image.

Images