Remote sensing image similarity calculation method based on co-located pixel adjacent domain feature vectors

Abstract

The invention discloses a remote sensing image similarity calculation method based on co-location pixel neighborhood domain feature vectors. The method comprises the following steps: firstly, comparing all pixels of remote sensing images with full-wave-band k-order numerical values of adjacent domain pixels, defining k-order adjacent domain full-wave-band feature vectors of the pixels, calculating a cosine distance between the k-order adjacent domain full-wave-band feature vectors of the pixels at the same geographic position of the two remote sensing images, and further calculating feature vector similarity of the same-position pixels; and finally calculating a similarity index of the two remote sensing images by multiplying a mean value of the similarity of all the pixel vectors by a proportion of the same geographic areas of the two images. On one hand, the spatial characteristics of the pixels are fully considered, and the requirements of geographic space big data analysis are better met; and on the other hand, by calculating the similarity of the adjacent domain feature vectors of the pixels at the same spatial position, the method disclosed by the invention is more suitable for similarity calculation of remote sensing images with the same wave band in the same region in different periods.

Description

Technical Field

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

[0002] Remote sensing image data acquired by various remote sensing platforms such as satellites, aviation and drones are not only important data sources for spatial big data, but also play an important role in geospatial big data analysis, spatial knowledge mining, and comprehensive application of spatial big data. With the continuous development and in-depth application of artificial intelligence methods and technologies, machine learning, deep learning and other methods have also played an increasingly important role in remote sensing image data processing. In the field of intelligent analysis and knowledge mining of remote sensing image data, similarity calculation is the basis of remote sensing image matching, machine cognition and other technologies. In recent years, with the continuous application of remote sensing image data processing technology and knowledge mining methods in the fields of ecological protection, land law enforcement, and farmland protection, calculating the degree of change of multi-period remote sensing image data is an important technical means to identify surface changes. The degree of change of remote sensing images is essentially to calculate the similarity of multi-period remote sensing images, and select remote sensing images with a similarity lower than a given threshold as key areas of concern, which can effectively reduce the manpower and material resources for surface change inspections, and also reduce the large amount of computing power required for remote sensing intelligent analysis to a certain extent.

[0003] After years of research, the similarity calculation methods of remote sensing image data are roughly divided into traditional visual comparison method, visual significant feature similarity method, similarity method based on semantic features and similarity method based on deep learning, etc. By analyzing the relevant research results of current remote sensing image data similarity calculation, it can be seen that the existing remote sensing image data similarity calculation methods are mainly measured based on the visual features, entity semantic features, convolution feature vectors and other aspects of the pixels of remote sensing image data, without considering the geographic spatial characteristics of the pixels of remote sensing image data, and ignoring the geographic variables of the corresponding geographic spatial position of the pixels and the surrounding areas.

[0004] Although the image processing software library OpenCV provides an image similarity calculation method based on pixel feature vectors and is widely used in image recognition, target detection, machine vision and other fields, the image similarity of OpenCV is measured based on the ratio of the number of key points of image matching to the total number of key points of the image, without considering the role of other non-key point pixels in image similarity. Therefore, this method is more suitable for image matching and stitching. In addition, the image processed by OpenCV is based on the raster data model of RGB primary colors, and when extracting feature points, the RGB model data needs to be converted into a grayscale model, ignoring the data characteristics of other bands of the image. 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 calculation formula of the correlation coefficient of numerical variables based on covariance, expand it and apply it to image similarity calculation. Although this method is simple in algorithm, it has a large amount of calculation and does not take into account the spatial characteristics of the pixels of the remote sensing image and the geographical variables of its 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 co-located pixels in the neighborhood to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above object, the present invention provides a remote sensing image similarity calculation method based on co-located pixel neighborhood feature vectors, 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 respectively, 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 feature vectors.

[0011] Preferably, obtaining the full-band k-order neighboring domain 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 the pixel value of its k-order neighboring domain, and obtain the feature vector according to 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 neighboring domain 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] In the formula, x i Represents the i-th component of the eigenvector of a remote sensing image pixel x, y i represents the i-th component of the feature vector of the same pixel y as 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 of 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 of two remote sensing images comprises:

[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 of two remote sensing images is:

[0030]

[0031] Where s1 and s2 represent the geographic spatial areas of the two remote sensing images, s overlay represents the area of ​​overlapping region, 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 on which a computer program is stored. When the computer program is executed by a processor, the steps of the method are implemented.

[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 of 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 proportion 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, on the one hand, 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 vector of the pixel at the same spatial position, the method disclosed in the present invention is more suitable for the similarity calculation of remote sensing images of the same band in the same region in different periods. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the 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 of ​​an embodiment of the present invention;

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

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

[0041] Figure 5 A similarity curve diagram of data on October 23, 2024 and other periods calculated by OpenCV in an embodiment of the present invention;

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

[0043] Figure 7 exemplifying diagrams 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 It is a numerical schematic diagram of three bands 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 in the embodiments of the present 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] The traditional visual comparison method is a method to judge the similarity of images based on human visual effects. It mainly uses the color, texture, brightness and other characteristics of remote sensing image data observed by the human eye, and combines domain knowledge for identification and judgment. This method not only requires interpreters to have rich domain knowledge, but also consumes a lot of manpower. Its disadvantage is that manual interpretation is highly subjective and it is difficult to objectively give similarity measurements. In addition, this method usually maps the invisible bands to the visible red, blue and green bands and synthesizes the images, which will cause the loss of invisible band data characteristics and cannot take into account the characteristics of all band data.

[0049] The visual salient feature similarity method imitates the visual principle of human observation of images and its image attention mechanism, and uses computer image data processing software or specific programs developed to enhance the features of the region of interest or entity, thereby obtaining the local salient features of the remote sensing image, and then calculating the similarity of the data. This method first needs to retrieve and define the region of interest or entity, and then enhance the color, brightness, saturation, etc. of the region of interest or entity in a targeted manner to obtain the salient features and geometric shape features of the observed subject. Therefore, this method is not only affected by the subjective factors of the selection of the region of interest or entity, but also requires the use of third-party software when enhancing and adjusting, and the setting of multiple parameters will also be restricted by many subjective factors, all of which affect the calculation of similarity to a certain extent.

[0050] The similarity method based on semantic features uses the semantic definition method to describe geographic entities and their relationships to describe the semantic features between entities in remote sensing images, and then calculate the data similarity. This method requires the use of remote sensing image segmentation and other technologies to identify the entities, and then establish a semantic feature model based on the spatial relationships such as co-location, inclusion, intersection, and proximity between entities. The repetition of remote sensing image entities is calculated based on the entities and their semantic features, and the repetition is used to measure the similarity of remote sensing images. This method combines multiple methods and technologies such as image segmentation, entity definition, and entity semantic description, and the processing process and calculation are relatively complicated.

[0051] The similarity method based on deep learning uses convolutional neural networks to train large-scale image samples to obtain an image feature recognition model, which is then used to calculate the features or encoding of the image, and then calculate the similarity of the image. In recent years, this method has been favored by a large number of researchers and technicians, and has been widely used in fields such as e-commerce. However, this method requires the training and learning of large-scale image samples, is suitable for similarity calculation of small-scale and small-scale images, and is mainly used for comparison of existing sample library images, not for similarity comparison of non-sample library images. For example, the camera search technology of e-commerce platforms for similar products is a successful application of this method. The similar product images searched by users are product images that have been trained and learned by the platform. This method is not suitable for remote sensing image data analysis with spatial information.

[0052] Embodiment 1

[0053] like Figure 1 As shown, in this embodiment, a remote sensing image similarity calculation method based on co-located pixel neighborhood feature vectors is provided, comprising 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 respectively, 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 feature vectors.

[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 of 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 number 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 line to line r p +k lines, starting from c p -k columns to c p +k columns, compare the values ​​of each pixel in the neighboring domain with the pixel p in turn. If the pixel value in the neighboring domain is greater than the pixel p, it is recorded as 1; if they are equal, it is recorded as 0; otherwise, it is recorded as -1. Thus, a 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 the 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 of the remote sensing image, including:

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

[0067]

[0068] In the formula, x i Represents the i-th component of the eigenvector of a remote sensing image pixel x, y i represents the i-th component of the feature vector of the same pixel y as 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] Among them, λ represents the similarity between the pixel x of R1 and the co-located pixel y of 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 spatial areas of the two images respectively.

[0078] In order to verify the technical effect of the present invention, the inventor selected an area of ​​about 9 km × 9 km in Changsha as the experimental area ( Figure 2 ), 12-band remote sensing image data of the Sentinel-2B satellite with a spatial resolution of 10m-60m and a cloud coverage rate of 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 inventor 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 the 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 on 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. 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 the key point matching method of OpenCV to calculate the similarity of images. Since OpenCV can only process three bands, red, blue, and green, this set of experiments uses the 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 image based on the red, blue and green bands of the remote sensing image, and then measure the image similarity based on the proportion of the matching number of key points, and cannot take into account all the band data; on the other hand, OpenCV only calculates the similarity from the matching number of 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 laws of real-world objects. In comparison, the calculation method of the present invention has a certain degree of conformity with the changing process of real-world objects, and the noise experiment results further demonstrate the reliability and availability of the method of the present invention.

[0086] Embodiment 2

[0087] This embodiment provides a remote sensing image similarity calculation method based on co-located pixel neighborhood feature vectors, 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 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 different, the similarity calculation ends; otherwise, the overlapping area of ​​the two images is calculated. If there is no overlapping area of ​​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 rule: From the rth p -k line to line r p +k lines, starting from c p -k columns to c p +k columns, compare the values ​​of each pixel in the neighboring domain with the pixel p in turn. If the pixel value in the neighboring domain is greater than the pixel p, it is recorded as 1; if they are equal, it is recorded as 0; otherwise, it is recorded as -1. Thus, a 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 the 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 second-order neighbors 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,-1,1,1,1,1,1). The vectors of the three bands are combined to form the neighborhood domain eigenvector 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,-1,1,0,1,-1,1,-1,1,1,1,1,1,1,1,1,1,1,-1,-1,-1,-1,0,0,0,1,-1,-1,0,1,-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 of the remote sensing image, including:

[0096] (3) Search for pixel y in remote sensing image R2 that has the same geographical location as pixel x in remote sensing image R1, and calculate the k-order neighborhood feature vector V of x and y in the entire band 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] Among them, λ represents the similarity between the pixel x of R1 and the co-located pixel y of R2.

[0101] For example, in the above example, the vector V of 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,-1,1,1,1,1,1,1,0,1,1,-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), assuming that the eigenvector V of the pixel y of image R2, which is co-located with x, is 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,-1,0,-1,0,1,-1,-1,-1,0,1,-1,-1,-1,0,1,-1,-1,-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 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 spatial areas of the two images respectively.

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

Claims

1. A remote sensing image similarity calculation method based on feature vectors of co-located pixels in the neighborhood, 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 respectively, 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 similarity of two remote sensing images is calculated based on the similarity of the full-band feature vectors.

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 the pixel value of its k-order neighboring domain, and obtain the feature vector according to 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 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 neighboring domain feature vectors of the pixels with the same geographical location in the two remote sensing images; The local similarity of pixels at the same geographical location in two remote sensing images is calculated based on the cosine distance.

4. The method according to claim 3, characterized in that The calculation expression of the cosine distance is: In the formula, x i Represents the i-th component of the eigenvector of a remote sensing image pixel x, y i represents the i-th component of the feature vector of the same pixel y as x in another remote sensing image, and n represents the number of features of the feature vector in the neighboring domain.

5. The method according to claim 3, characterized in that: 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.

6. The method according to claim 1, characterized in that The calculating the similarity of two remote sensing images comprises: 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.

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

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. 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 7 are implemented.

Images