A remote sensing image similarity calculation method based on feature vectors of co-located pixels' neighborhood domain

Abstract

The present invention discloses a method for calculating the similarity of remote sensing images based on the neighborhood feature vectors of co-located pixels. The present invention first compares the full-band k-order values ​​of all pixels of a remote sensing image with those of its neighborhood pixels, defines the full-band k-order neighborhood feature vectors of pixels, calculates the cosine distance of the full-band k-order neighborhood feature vectors of pixels of the same geographical location in two remote sensing images, and then calculates the similarity of the feature vectors of the co-located pixels; finally, the mean of the similarities of all pixel vectors is calculated and multiplied by the ratio of the same geographical area of ​​the two images to obtain the similarity index of the two remote sensing images. On the one hand, the present invention fully takes into account the spatial characteristics of pixels and is more in line with the requirements of geospatial big data analysis; on the other hand, by calculating the similarity of the neighborhood feature vectors of pixels of the same spatial position, the method disclosed by the present invention is more suitable for similarity calculation of remote sensing images of the same band in the same region in different periods.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a remote sensing image similarity calculation method based on feature vectors of a neighborhood domain of co-located pixels. Background Art

[0002] Remote sensing image data acquired by various remote sensing platforms, including satellites, aerial vehicles, and drones, is not only an important source of spatial big data but also plays a crucial role in geospatial big data analysis, spatial knowledge mining, and the integrated application of spatial big data. With the continuous development and in-depth application of artificial intelligence methods and technologies, machine learning and deep learning are playing an increasingly important role in remote sensing image data processing. In the intelligent analysis and knowledge mining of remote sensing image data, similarity calculation forms the foundation of remote sensing image matching and machine cognition. In recent years, with the increasing application of remote sensing image data processing technologies and knowledge mining methods in fields such as ecological protection, land law enforcement, and farmland protection, calculating the degree of change in multi-period remote sensing image data has become an important technical means for identifying surface changes. Remote sensing image change essentially involves calculating the similarity of multiple remote sensing images. Remote sensing image change, then, selects images with similarity below a given threshold as key areas of focus. This effectively reduces the manpower and material resources required for surface change inspections and, to a certain extent, reduces the significant computing power required for intelligent remote sensing analysis.

[0003] After years of research, methods for calculating remote sensing image data similarity generally include traditional visual comparison methods, visual salient feature similarity methods, semantic feature-based similarity methods, and deep learning-based similarity methods. Analysis of current research results on remote sensing image data similarity calculation reveals that existing methods primarily measure remote sensing image data similarity based on the visual features, entity semantic features, and convolutional feature vectors of remote sensing image data pixels. These methods fail to consider the geographic spatial characteristics of remote sensing image data pixels and ignore the geographic variables of the pixel's corresponding geographic spatial location and the surrounding area.

[0004] Although the image processing software library OpenCV provides a method for calculating image similarity based on pixel feature vectors and has been widely used in fields such as image recognition, object detection, and machine vision, OpenCV's image similarity is measured based on the ratio of the number of matching key points to the total number of key points in the image. This method does not take into account the role of non-keypoint pixels in image similarity, making it more suitable for image matching and stitching. Furthermore, images processed by OpenCV are based on a raster data model based on the RGB primary colors. When extracting feature points, the RGB model data must be converted to a grayscale model, ignoring the data characteristics of other image bands. Therefore, this method is not suitable for similarity calculation of multispectral or hyperspectral remote sensing images.

[0005] In addition, some image similarity methods improve the formula for calculating the correlation coefficient of numerical variables based on covariance, expand it, and apply it to image similarity calculation. Although this method is simple, it is computationally intensive and does not take into account the spatial characteristics of the pixels in the remote sensing image and the geographic variables of their adjacent pixels. Therefore, this method is not suitable for multi-band remote sensing image data. Summary of the Invention

[0006] The present invention proposes a remote sensing image similarity calculation method based on feature vectors of neighboring domains of co-located pixels to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above object, the present invention provides a method for calculating remote sensing image similarity based on feature vectors of co-located pixel neighborhoods, comprising the following steps:

[0008] Read two remote sensing images, calculate the k-order neighborhood feature vector of each pixel in the two remote sensing images, and obtain the full-band k-order neighborhood feature vector;

[0009] Calculate the similarity of the full-band feature vector of each pixel in the same geographical location of two remote sensing images;

[0010] The similarity of two remote sensing images is calculated based on the similarity of the full-band eigenvectors.

[0011] Preferably, obtaining the full-band k-order neighborhood feature vector includes:

[0012] Search the k-order neighboring domain of pixel p in a band data of a remote sensing image;

[0013] Compare the pixel p with its k-order neighboring pixel values, and obtain the feature vector based on the comparison result;

[0014] Iterate the above steps to obtain the feature vectors of all bands;

[0015] In the order of band numbers from small to large, all band feature vectors of the same pixel are connected and merged to form the full-band k-order neighborhood feature vector of the pixel.

[0016] Preferably, the calculating of the similarity of the full-band feature vectors of each pixel at the same geographical location of two remote sensing images comprises:

[0017] Obtain pixels with the same geographical location in the remote sensing images, and calculate the cosine distance of the full-band k-order neighborhood feature vectors of the pixels with the same geographical location in the two remote sensing images;

[0018] The local similarity of pixels at the same geographical location in two remote sensing images is calculated based on the cosine distance.

[0019] Preferably, the calculation expression of the cosine distance is:

[0020]

[0021] Where x i Represents the i-th component of the eigenvector of a remote sensing image pixel x, y i It represents the i-th component of the feature vector of the pixel y that is co-located with x in another remote sensing image, and n represents the number of features of the feature vector in the neighboring domain.

[0022] Preferably, the expression for calculating the local similarity of pixels at the same geographical location in two remote sensing images is:

[0023] λ = 1 - θ / 2;

[0024] Where θ represents the cosine distance.

[0025] Preferably, the calculating the similarity between the two remote sensing images includes:

[0026] Calculate the overlapping area of ​​the two images;

[0027] Calculate the average local similarity of all pixels in the overlapping area;

[0028] The similarity of the two remote sensing images is calculated based on the average value of the overlapping area and the local similarity of the pixels.

[0029] Preferably, the expression for calculating the similarity between two remote sensing images is:

[0030]

[0031] Where s1 and s2 represent the geographic space areas of two remote sensing images, s overlay represents the area of ​​overlapping regions, and θ represents the average value of local similarity of pixels.

[0032] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0033] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0034] Compared with the prior art, the present invention has the following advantages and technical effects:

[0035] The remote sensing image similarity calculation method disclosed in the present invention fully considers the spatial position relationship of pixels, as well as the size of the pixel and its k-order neighboring domain pixel value, establishes the full-band neighboring domain feature vector of the pixel, and then uses the cosine distance of the feature vector of the pixel at the same geographic spatial position in the remote sensing image to define the local similarity of the pixel, and finally calculates the average value of the similarity of all pixels in the same geographic spatial region multiplied by the ratio of the area of ​​the same geographic spatial region to the area of ​​the remote sensing image to obtain the similarity of the remote sensing image. Compared with the existing image data similarity calculation method, the similarity method disclosed in the present invention fully takes into account the spatial characteristics of the pixel and is more in line with the requirements of geospatial big data analysis; on the other hand, by calculating the similarity of the neighboring domain feature vectors of the pixels at the same spatial position, the method disclosed in the present invention is more suitable for similarity calculation of remote sensing images of the same band in the same region in different periods. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0037] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of an experimental area according to an embodiment of the present invention;

[0039] Figure 3 This is a similarity curve diagram of remote sensing image data on October 23, 2024, and other periods according to an embodiment of the present invention;

[0040] Figure 4 A noise ratio and similarity curve diagram according to an embodiment of the present invention;

[0041] Figure 5 This is a similarity curve chart of the data on October 23, 2024, calculated by OpenCV in an embodiment of the present invention, with respect to the data of other periods;

[0042] Figure 6 A similarity curve graph of October 23, 2024, calculated by OpenCV in an embodiment of the present invention and its noise-added data;

[0043] Figure 7 1 is an example diagram of the first-order, second-order, and third-order neighborhoods of a pixel p according to an embodiment of the present invention;

[0044] Figure 8 Schematic diagram of three band values ​​of a pixel and its second-order neighboring domain according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0046] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0047] The existing methods are introduced as follows:

[0048] Traditional visual comparison methods rely on human visual perception to determine image similarity. They primarily utilize the human eye to observe features such as color, texture, and brightness in remote sensing image data, combined with domain knowledge for identification and judgment. This method not only requires interpreters to possess extensive domain knowledge but also consumes significant manpower. A drawback is that manual interpretation is highly subjective, making it difficult to objectively measure similarity. Furthermore, this method typically involves mapping invisible bands into visible red, blue, and green bands for image synthesis, which results in a loss of data features in the invisible bands and fails to account for the characteristics of all bands.

[0049] The visual salient feature similarity method, modeled on the principles of human visual observation and image attention mechanisms, utilizes computer image data processing software or specialized programs to enhance features of regions of interest (ROIs) or entities, thereby capturing local salient features in remote sensing imagery and then calculating data similarity. This method first requires retrieving and defining ROIs or entities, then specifically enhancing their color, brightness, saturation, and other attributes to capture the salient features and geometric shape characteristics of the observed subject. Therefore, this method is not only subject to the subjective factors of ROI or entity selection, but also requires the use of third-party software for enhancement and adjustment. The setting of various parameters is also subject to numerous subjective factors, all of which affect the similarity calculation to a certain extent.

[0050] Similarity methods based on semantic features use semantic definition methods to describe geographic entities and their relationships, describing the semantic features between entities in remote sensing images and then calculating data similarity. This method requires the use of techniques such as remote sensing image segmentation to identify entities. It then establishes a semantic feature model based on spatial relationships between entities, such as colocation, inclusion, intersection, and proximity. The repetition of remote sensing image entities is calculated based on the entities and their semantic features, thus using repetition to measure remote sensing image similarity. This method combines multiple methods and techniques, including image segmentation, entity definition, and entity semantic description, resulting in a complex processing and computational process.

[0051] Deep learning-based similarity methods utilize convolutional neural networks to train large-scale image samples to generate image feature recognition models. This model is then used to calculate image features or encodings, and then to calculate image similarity. In recent years, this method has gained widespread favor among researchers and technicians, and has been widely applied in fields such as e-commerce. However, this method requires training and learning from large-scale image samples, making it suitable for similarity calculations on small-scale and small-scale images. Furthermore, it is primarily used for comparing images within a pre-existing sample library, and is not suitable for similarity comparisons of images outside of a sample library. For example, the photo search technology used by e-commerce platforms is a successful application of this method, where users search for similar product images based on product images that have been trained and learned by the platform. This method is not suitable for analyzing remote sensing imagery data containing spatial information.

[0052] Example 1

[0053] like Figure 1 As shown, this embodiment provides a remote sensing image similarity calculation method based on the feature vector of the neighboring domain of the same-position pixel, including the following steps:

[0054] Read two remote sensing images, calculate the k-order neighborhood feature vector of each pixel in the two remote sensing images, and obtain the full-band k-order neighborhood feature vector;

[0055] Calculate the similarity of the full-band feature vector of each pixel in the same geographical location of two remote sensing images;

[0056] The similarity of two remote sensing images is calculated based on the similarity of the full-band eigenvectors.

[0057] The specific steps are as follows:

[0058] Step 1: Read two remote sensing images and calculate the k-order neighborhood feature vector of each pixel, including:

[0059] (1) In a band of data in a remote sensing image, search for the k-order neighborhood N of pixel p. The k-order neighborhood is a set of pixels whose row or column distance from p is less than or equal to k, which can be expressed as:

[0060] N={q||r p -r q |≤kv|c p -c q |≤k}

[0061] Where q represents the neighboring pixel, r p and r q Represents the row numbers of p and q pixels respectively, c p and c q Represents the column number of p and q pixels respectively. For example, Figure 7a~7c give examples of the 1st, 2nd, and 3rd order neighborhoods of pixel p.

[0062] (2) Compare the pixel p with its k-th neighboring pixel values ​​to form a feature matrix. The comparison rule is: p -k lines to line r p +k lines, starting from c p -k columns to c p +k columns, compare the value of each pixel in the neighboring domain with the value of pixel p in turn. If the value of the pixel in the neighboring domain is greater than the value of pixel p, it is recorded as 1; if they are equal, it is recorded as 0; otherwise it is recorded as -1. Thus, the feature vector of a band is obtained, with a dimension of (1+2k) 2 .

[0063] (3) Repeat process (1)-(3) to calculate the eigenvectors of all other bands.

[0064] (4) In order of band numbers from small to large, all band feature vectors of the same pixel are connected and merged to form the full-band k-order neighborhood feature vector of the pixel, with a dimension of n = b(1+2k) 2 , where b is the number of bands of the remote sensing image.

[0065] Step 2: Calculate the similarity of the full-band feature vector of each pixel in the remote sensing image, including:

[0066] (1) Search for pixel y in the remote sensing image R2 that has the same geographical location as pixel x in the remote sensing image R1, and calculate the k-order neighborhood feature vector V of the full band of x and y x =(x1,x2,...,x n ), V y =(y1,y2,...,y n ) is calculated as follows:

[0067]

[0068] Where x i Represents the i-th component of the eigenvector of a remote sensing image pixel x, y i It represents the i-th component of the feature vector of the pixel y that is co-located with x in another remote sensing image, and n represents the number of features of the feature vector in the neighboring domain.

[0069] (2) Calculate the local similarity between R1 pixel x and R2 pixel y. The calculation formula is expressed as:

[0070] λ=1-θ / 2

[0071] Here, λ represents the similarity between pixel x in R1 and its co-located pixel y in R2.

[0072] Step 3: Calculate the similarity of remote sensing images. The process includes:

[0073] (1) Calculate the overlapping area s of the two images overlay ;

[0074] (2) Calculate the average local similarity of all pixels in the overlapping area

[0075] (3) Use the following formula to calculate the similarity between R1 and R2:

[0076]

[0077] Among them, s1 and s2 represent the geographic space areas of the two images respectively.

[0078] In order to verify the technical effect of the present invention, the inventors selected an area of ​​about 9 km × 9 km in Changsha as the experimental area ( Figure 2 ), remote sensing image data of the Sentinel-2B satellite with 12 bands and a spatial resolution of 10m to 60m with cloud coverage less than 5% were selected from the website https: / / browser.dataspace.copernicus.eu / . The data were collected on October 23, 2024, September 23, 2024, August 4, 2024, December 28, 2023, September 14, 2022, March 28, 2022, December 5, 2021, August 30, 2021, November 13, 2020, February 17, 2020, November 19, 2019, November 24, 2018, and April 18, 2018, for a total of 13 data periods. In order to normalize the spatial resolution of remote sensing images, the inventors resampled the data of each band in the same period and finally stored them as image data with a spatial resolution of 10m.

[0079] The inventors conducted two sets of experiments using the above data:

[0080] Experiment I. This group of experiments uses the calculation method of the present invention to calculate the similarity between the data on October 23, 2024 and the data of other periods, and the similarity curve is drawn as follows Figure 3 shown.

[0081] from Figure 3 It can be seen that the longer the time is from October 23, 2024, the lower the similarity of remote sensing images is. This is basically consistent with the degree of image change caused by urban development and construction, which also shows that the method of the present invention has good practicality.

[0082] In order to further verify the feasibility of the method of the present invention, the inventors added a certain proportion of salt and pepper noise based on the data of October 23, 2024, and then calculated the similarity with the original data to obtain the noise ratio and similarity curve as shown in the figure below: Figure 4 shown.

[0083] from Figure 4 It can be seen that as the noise ratio increases, the image similarity decreases, which further proves the feasibility and usability of the method of the present invention.

[0084] Experiment II. This set of experiments uses OpenCV's keypoint matching method to calculate image similarity. Since OpenCV can only process red, blue, and green bands, this set of experiments uses these three bands of the experimental data for similarity calculation. Figure 5 The similarity curve between the data on October 23, 2024 and the data of other periods is shown. Figure 6 The similarity curve between the data on October 23, 2024 and the data with a certain proportion of salt and pepper noise added is shown.

[0085] Comparing the two sets of experimental results, it can be seen that OpenCV can only calculate the key points of the remote sensing image based on the red, blue, and green bands, and then measure the image similarity based on the proportion of key point matches, which cannot take into account the data of all bands. On the other hand, OpenCV only calculates similarity based on the number of matches of the key point features of the image, ignoring the spatial co-location relationship of the pixels. Therefore, the image similarity calculation results of OpenCV fluctuate greatly and deviate from the changing patterns of real-world objects. In comparison, the calculation method of the present invention has a certain degree of consistency with the changing process of real-world objects. The noise experiment results further demonstrate the reliability and usability of the method of the present invention.

[0086] Example 2

[0087] This embodiment provides a method for calculating remote sensing image similarity based on feature vectors of co-located pixel neighborhoods, including the following steps:

[0088] Step 1: Read two phases of remote sensing image data and calculate the k-order neighborhood feature vector of each pixel, including:

[0089] (1) Load and read the remote sensing image data R1 and R2, and determine whether the spatial resolution and number of bands of the two images are the same. If they are not, the similarity calculation ends; otherwise, the geographic overlap area of ​​the two images is calculated. If there is no geographic overlap area between the two images, the similarity calculation ends.

[0090] (2) In each of the two images, search for the k-order nearest neighbor pixels of each pixel in turn and calculate the feature vector of each pixel, including:

[0091] Compare the pixel p with its k-th neighboring pixel values, and the comparison results form a feature matrix. Comparison rules: From the rth p -k lines to line r p +k lines, starting from c p -k columns to c p +k columns, compare the value of each pixel in the neighboring domain with the value of pixel p in turn. If the value of the pixel in the neighboring domain is greater than the value of pixel p, it is recorded as 1; if they are equal, it is recorded as 0; otherwise it is recorded as -1. Thus, the feature vector of a band is obtained, with a dimension of (1+2k) 2 .

[0092] Repeat the above steps to calculate the eigenvectors of all bands.

[0093] In order of band numbers from small to large, all band feature vectors of the same pixel are connected and merged to form the full-band k-order neighborhood feature vector of the pixel, with a dimension of n=b(1+2k) 2 , where b is the number of bands of the remote sensing image.

[0094] For example, if there is a 3-band remote sensing image data, the 2nd order neighbor of a pixel and its band values ​​are as follows: Figure 8 As shown, the neighborhood vector of the first band is (-1,1,1,-1,-1,1,1,1,1,-1,1,0,1,-1,-1,1,1,-1,1,1,-1,-1,-1), the neighborhood vector of the second band is (1,1,1,1,1,1,1,0,1,1,-1,1,0,1,-1,1,-1,1,1,1,1,1,1,1), and the neighborhood vector of the third band is (-1,-1,-1,-1,-1,0,0,0,1,-1,-1,-1,0,1,-1,-1,-1,-1,1,-1,1,1, The vectors of the three bands are combined to form the neighborhood feature vector of the pixel: (-1,1,1,-1,-1,1,1,1,1,-1,1,0,1,-1,-1,1,1,-1,1,1,1,-1,-1,-1,1,1,1,1,1,0,1,1,-1,1,-1,1,1,1,1,1,1,1,1,1,-1,-1,-1,-1,0,0,0,1,-1,-1,-1,0,1,-1,-1,-1,-1,-1,1,1,1,0,1,-1).

[0095] Step 2: Calculate the similarity of the full-band feature vector of each pixel in the remote sensing image, including:

[0096] (3) Search for pixel y in the remote sensing image R2 that has the same geographical location as pixel x in the remote sensing image R1, and calculate the k-order neighborhood feature vector V of the full band of x and y x =(x1,x2,...,x n ), V y =(y1,y2,...,y n ) is calculated as follows:

[0097]

[0098] (4) Calculate the local similarity between R1 pixel x and R2 pixel y. The calculation formula is expressed as:

[0099] λ=1-θ / 2

[0100] Here, λ represents the similarity between pixel x in R1 and its co-located pixel y in R2.

[0101] For example, in the above example, the vector V of the pixel x of image R1 is x =(-1,1,1,-1,-1,1,1,1,1,-1,1,0,1,-1,-1,1,1,-1,1,1,1,-1,-1,-1,-1,1,1,1,1,1,1,0,1,1,-1,1,-1,-1,1,1,1,1,1,1,1,-1,-1,-1,-1,0,0,0,1,-1,-1,-1,0,1,-1,-1,-1,-1,1,-1,1,1,0,1,-1), assuming that the eigenvector V of the pixel y of the image R2 is collocated with x y =(-1,0,1,-1,-1,1,0,0,0,1,-1,1,0,1,-1,-1,0,1,-1,-1,-1,1,1,1,1,1,1,0,1,1,-1,1,0,1,-1,-1,-1,1,0,0,1,1,1,1,-1,-1,-1,-1,0,-1,0,1,-1,-1,-1,0,1,-1,-1,-1,-1,1,1,0,1,-1,-1,-1,1,1,0,1,-1), the cosine distance between two pixels θ = 0.06945, and the local similarity λ = 0.9653.

[0102] Step 3: Calculate the similarity between the two remote sensing images. The process includes:

[0103] (1) Calculate the overlapping area s of the two images overlay ;

[0104] (2) Calculate the mean of the local similarity of all pixels in the overlapping area

[0105] (3) Use the following formula to calculate the similarity between R1 and R2:

[0106]

[0107] Among them, s1 and s2 represent the geographic space areas of the two images respectively.

[0108] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A remote sensing image similarity calculation method based on feature vectors of co-located pixel neighborhoods, characterized in that: The following steps are involved: Read two remote sensing images, calculate the k-order neighborhood feature vector of each pixel in the two remote sensing images, and obtain the full-band k-order neighborhood feature vector; Calculate the similarity of the full-band feature vector of each pixel in the same geographical location of two remote sensing images; The calculation of the full-band feature vector similarity of each pixel in the same geographical location of two remote sensing images includes: Obtain pixels with the same geographical location in the remote sensing images, and calculate the cosine distance of the full-band k-order neighborhood feature vectors of the pixels with the same geographical location in the two remote sensing images; Calculating the local similarity of pixels at the same geographical location in two remote sensing images according to the cosine distance; The expression for calculating the local similarity of pixels of the same geographical location in two remote sensing images is: λ = 1-θ / 2; Where θ represents the cosine distance; Calculate the similarity of two remote sensing images based on the similarity of full-band eigenvectors; The similarity calculation of two remote sensing images includes: Calculate the overlapping area of ​​the two images; Calculate the average local similarity of all pixels in the overlapping area; The similarity of the two remote sensing images is calculated based on the average value of the overlapping area and the local similarity of the pixels.

2. The method according to claim 1, characterized in that Obtaining the full-band k-order neighboring domain feature vector includes: Search the k-order neighboring domain of pixel p in a band data of a remote sensing image; Compare the pixel p with its k-order neighboring pixel values, and obtain the feature vector based on the comparison result; Iterate the above steps to obtain the feature vectors of all bands; In the order of band numbers from small to large, all band feature vectors of the same pixel are connected and merged to form the full-band k-order neighborhood feature vector of the pixel.

3. The method according to claim 1, characterized in that The calculation expression of the cosine distance is: Where x i Represents the i-th component of the eigenvector of a remote sensing image pixel x, y i It represents the i-th component of the feature vector of the pixel y that is co-located with x in another remote sensing image, and n represents the number of features of the feature vector in the neighboring domain.

4. The method according to claim 1, wherein The expression for calculating the similarity between two remote sensing images is: Where s1 and s2 represent the geographic space areas of the two remote sensing images, s overlay represents the overlapping area, It represents the average value of local similarity of pixels.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images